UNLSSIFIED Eh h hh h. PACKET RADIO APR DEPLOYMENT … · B. Scenario Generator Module As depicted...
Transcript of UNLSSIFIED Eh h hh h. PACKET RADIO APR DEPLOYMENT … · B. Scenario Generator Module As depicted...
AD-AC87 417 NETWORK ANALYSIS CORP GREAT NECK NY F/S 17/2.1PACKET RADIO DEPLOYMENT STUDY. CU)APR 60 DAAKSO-79-C-0763
UNLSSIFIED FR.207.01-Ri L
.3CL Eh h hh h.
1111111112 .211111 _111112-0
IIIII ,.o . I II,8
111125 1.6
MICR ;OPY RESOLUTION TE ,. ,,.....J
44
Kv
iv2e Ail.
7Z
0 44
"Af
P
lz
lyva "toy,IWO
S
4 04A,
I P
i is
-4434"
(P-4-ji
91,27-4-EV,
2, 3
AM,
V
LEM,~
'11JNc rX -~~J -I' -C 6
PACKETJADIO DEPLOYMENT STUDY
PERFOR MED FOR:
USA CORADCOM - IUDRDCO-COM-RF-2
Fort Monmouth, NJ 07703
April 10, 1980
NETWORK ANALYSIS CORPORATION/130 Steamboat Road '
Great Neck, NY 11024
Apprvedforpubi~c re4e;
* ii CO-TABLE OF CONTENTS
1. INTRODUCTION . . . ........ ....................... .. .. .. .. .......
2. MILITARY SCENARIO MODEL .......................... 2.1
2.1 Overview ......... .... ... .... .... ... ...... 2.1
2.2 Data Base . . ......... ... .... ... .... ........ 2.8
2.3 Scenario Generator .. .. .... ..... ...... ..... 2.372.4 The Link Module. .. .. ...... ...... ..... ... 2.422.5 Packet Radio Simulation Module. .. ... ...... ...... 2.44
3.~ SIMULATION SCENARIOS .. .. ..... ..... ...... ... 3.1
3.1 Simulation Variables .. .. .. ..... ..... ...... .. 3.13.2 Statistics . . . ......... .... .... ... .... ..... 3.12
4. SIMULATION RESULTS .. .......................... 4.1
4.1 Summary . . ......... .... ... .... .... ....... 4.2
*4.2 Simulation Results of the Brigade Scenario. .. .. .. ....... 4.34.3 Simulation of 32PRU's Net Scenario .. .. .. .... ....... 4.114.4 Simulation Results of Support Scenario with either
Host/Poit-ointRotnt ou.t..n. . . .. .. . .. . . .. 4.17
5. THROUGHPUT ANALYSIS FOR TANDEM REPEATER NETWORK . . . . . 5.1
5.1 Summary and ntroduction .. . . .. . .. . . . .. .. . .. . 5.1
*5.2 Perfect Scheduling Analysis . . . . . . . . . . . . . . . . . . . 5.4
5.3 Tandem Repeaters U uing gCSA... . .. .. .. .. .. . . .. 5.65.4 Tandem Repeaters Using the ALOHA Aces Method . . . . . . . . 5.105.5 Tandom Network Using CSMA (Exat Methou) . . . . . . . . . . . 5.13
*5.6 Application: A Loose Bound onPerformance . . . . . . . . . . . . 5.23
n1ac.TABLE OF CONTENTS
6. SIJRVIVABILITY MODELS .. . . . . . . .. .. . .. . . . . . . . . 6.1
6.2 Conclusions . . . . . . .. .. .. .. . .. . . . . . . . . .6.6
7. ROUTING .. . . . . . .. .. .. . .. . . . .. .. .. .. . .7.1
7.1 Overview. .. .. ..... ..... .. .. .. .. .. ... 7.1
7.2 Motivation . . . . . . . . .. .. .. .. . .. . . . . . ... 7.2
7.3 Routing Procedure .. .. ... ..... ...... ....... 7.30 7.4 Computational Experience. .. . .. . . . .. .. .. .. .. 7.6
8. REPEATER LOCATION ALGORITHM . ... .. .. . .. ... .. .. 8.1
8.1 Introduction . . . .. .. .. . .. ... . .. .. .. .. .. 8.1
8.2 Set Covering Problem . .. .. .. .. .. .. .. .. .. .. 8.2
APPENDIX A: LONGLEY RICE CONNECTI; ITY ANALYSIS . . . . .. .. . .A.1
APPENDIX B: FEASIBILITY OF MOBILE SUBSCRIB3ER ENTRY SYSTEM ANDPACKET RADIO SYSTEM TO SUPPORT ARMY DATA NEEDLINES. B.1
Access on For
fiI R&
INTRODUCTION AND SUMMARY
ihs reot documents the results ,of the Packet Radio Deployment Study peformed)Network Analysis Corporation (NAC) for the Commn/utr-L-----e and Development
Command (CORADCOM) at Fort Monmouth. The jeetiveNof is to assess the
potential of the Packet Radio technology for satisfying communications requirements in the
military environment. The focus 6f the study has been on quantifying PR network
performance under the expected military traffic
NAC's activities have consisted o he following three tasks:-
- ,Task-it Enhancements ofN*e%_Packet Radio Simulation Progrh,(PRSM)
- Task 2. Implementation of a Military Scenario Model,
- Task -9 Design, Routing and Performance Analysis of Packet Radio Networks
for the Military Scenarios.
In Task 1, NAC modified PRSIM to more accurately model the capabilities of the
Defense Advanced Research Projects Agency's (DARPA) Test Bed PR Network. The
modifications of the NAC model Included:
- Integral terminal relay capability
- Point-to-point routing
C - Enhanced terminal queue management (to handle multiple TCP connections
through a single origination or destination PRU)
- Local repeater on packet transmission/reception
- Enhanced statistics gathering capability
- Mobile terminal capability.
1.1
Ia .C,Task 2 involved the development of a computerized data base and software modules
for Scenario Generation. To assist NAC in the development of these modules, MITRE
(tasked by CORADCOM) prepared a detailed description of a military scenario[ 11 This
description was based on the deployment of packet radio units in the Fulda Gap area of
Germany as part of the U.S. Army's Scores il-A scenario. The activities in this task
consisted of:
- Computerizing the terminal location, traffic volume, and mobility tracks
supplied by MITRE
- Developing software to transform this information into a form suitable for
PRSIM
- Modifying PRSIM to accept this information.
The detailed procedures for performing this task are described in Section 2 below.
In addition, the Longley Rice propagation model [6] was implemented and PRSIM was
modified to access this model to determine network connectivity. Subsequently, PRSIM was
modified to accept line of sight information to determine network connectivity.
Task 3 involved detailed analysis of the PR networks using both analytic models and
simulation. The issues addressed consisted of throughput-delay, survivability, routing, and
repeater location.
In this task, NAC flit performed a detailed analysis of network connectivity using the
Longley-Rice model. The results of these studies are described in Appendix A. Because of
the uncertainty of this model, it was decided to employ a LOS criteria to determine network
connectivity. Using the LOS criteria to determine network connectivity, NAC performed
extensive simulation to evaluate performance of a Command and Control Network under a
wide range of parameters. The range of these experiments are described in Section 3 and
the corresponding results in Section 4. The major conclusions are:
C2 traffic load can be handled by PR nets over a wide range of operating
conditions.
1.2
-na._Mean round trip (end-to-end) mesage delays of less than 2 seconds in the worst
case were measured. This included the increase in traffic of 100% over the
estimated volume and the reduction of PRU buffers from a nominel value of 6 to
3.
These conclusions hold for both host and direct (terminal-to-terminal) routing
although the performance of point-to-point routing is superior.
The introduction of two dedicated repeaters in the forward area near the
brigades decreases the number of hops reqWred on certain routes with host
routing; hence, message and packet delays are also decreased.
The retrograde from the initial deployment to Defense Bravo reduces network
connectivity, but delay requirements are still satisfied.
- Buffer size was identified as a sensitive parameter.
In addition to the simulation performance models, NAC also developed analytic models
for network throughput and survivability, these models are described in Sections 5 and 6. In
the throughput models, the performance of tandem networks was first addressed. The
objective was to estimate the throughput level such a network could support; tandem
networks were analyzed because initial connectivity studies indicated that PR networks
would be sparsely conneted. These results were subsequently applied to a general topology
to obtain a lower bound on the throughput level a network could support. The resulting
bound reinforced the simulation results, indicating that the network could very likely
support the offered load in the Command and Control network.
For the survivability problem, tandem networks were again first addressed because the
Initial Longley-Rie models indicated that the network connectivity would be sparse. Three
models were developed to estimate the number of repeaters required to guarantee a
specified survivability level However, with the LOS connectivity criteria, the PR networks
were much more densely connected, and the survivability problem was much lea complex.
In this case, for the C2 network certain Maneuver Battalions were only connected to the
remainder of the network by a single repeater. However, with the introduction of two
repeaters (as selected by MITRE), all Maneuver Battalions were 3-connected, which
guaranteed an adequate survivability level.
0 1.3
i .......... n a.. _A heuristic algorithm, described in Section 7, was developed to assign routes to the
needlins. The procedure is static in that routes are assigned a priori, independent of
network conditions. This procedure was developed to select shortest path routes with
minimal congestion, and was used in the simulations with connectivity determined by the
Longley-Rice criteria.
The configuration issued addressed focused on the location of repeaters. With the
topology defined by the LOS criteria, the problem was trivial, good repeater locations could
be "eye balled." However, with other criteria, the configuration problem is more
interesting. A procedure for selecting repeater locations is described in Section 8.
Also, as part of this study, an evaluation of the Mobile Subscriber Entry system was
performed. The results of this study were presented originally to DARPA and CORADCOM
in November 1978. The slides presented at that time are included in Appendix B.
C
1.4
2. MIUTARY SCENARIO MODEL
2.1 Overview
In this section the military scenario model and the methodology for carrying out its
analysis are described. In order to perform the design and analysis of the packet radio
networks proposed for deployment in the military scenarios, the following software modules
are required:
- Data base module
- Scenario generator module
- Packet radio simulator module
- Communications link module
- Connectivity module
- Configuration and routing module.
The interaction of these modules is schematically depicted in Figure 2.1. A solid line from
module A to module B (direction denoted by an arrow) indicates that module A accesses
module B as module A runs; a dashed line from module B to module A indicates that module
B generates an input file for module A. The functions of the individual modules are
C summarized below.
A. Data Base Module
The Data Base module will consist of location, traffic, and mobility files defining the
military scenarios. The information included in these files is derived from the military
scenarios defined by MITRE [1] with only minimal algebraic transformations performed for
convenience. This module, which is described in detail in Section 2.2, was developed as part
c of Task 2.
2.1
14BLITYTAFICTRMNLj
LINK CONFIGURATIONSCENARIO SCENARIO MDLAND
SUBNET GENERATOR IROUTING PROG.
SNAPSHOTCONNECTIVITY
! II I
I I
CONNECTEDI1 SUBSETS
TERMINAL TRAFFIC AND1
NOBILITY PROFILES II
I I
f" CONFIGURATIONANROTG
PRSI14
SCENARIOANALYSIS
FIGURE 2.1: OME INTERRELATIONSHIP
I0"
B. Scenario Generator Module
As depicted in Figure 2.1, the Scenario Generator Module accesses the location,
traffic, and mobility files in the Data Base Module and then generates input files for thepacket radio simulation module. The major parameters describing a scenario are:
- Subnetwork
Snapshot
Time duration.
In this study, only the Command and Control subnetwork was simulated although data bases
were created for Field Artillery and Air Defense applications.
As combat conditions dictate, the PR units will have to be reorganized; hence, threesnapshots were defined. A snapshot defines a particular deployment of the packet radio
units for the specified subnetworks; the following deployments have been defined:
- Initial D day H hour deployment
- D day H+6 hours deployment along Defense Line (DL) ALPHA
- D day H+24 hours deployment along Defense Line (DL) BRAVO.
Only the initial deployment and DL Bravo snapshots were simulated.
The Scenario Generator has the capability of generating a traffic profile for trafficrouted directly terminal-to-terminal or via a message-switching host. The time duration
input parameter defines the length of time covered by the scenario. The needlines havebeen defined on a terminal basis for each subnetwork, independent of deployment. However,adjustments to the traffic flows are made to account for terminals that are lost, captured,
or destroyed between subsequent scenarios.
After a terminal has been lost, destroyed, or captured in a particular deployment, theterminal i not included in subsequent deployments (and hence, obviously will not transmit
any mesages). However, in the subsequent deployments it is assumed that all terminals
2,3
0
that transmit are aware of all potential destination terminals that have been lost, captured,
or destroyed; hence, terminals do not attempt to transmit to these terminals that have been
lost due to attrition.
The output of the Scenario Generator consists of the following files to be used as input
for the packet radio simulation module:
- A PR locations file specifying coordinates of each terminal PR in the subnetwork
being considered.
- A traffic file specifying the times of transmissions, the transmitting PR,
receiving PR, and other message parameters.
- Mobile PR locations as a function of time.
C. Packet Radio Simulation Module
The Packet Radio Simulation Module is based on NAC's existing packet radio simulator
PRSIM. This program was extensively modified to reflect more accurately the PR test bed
protocols (as part of Task 1) and to accept data for the military scenario (as part of Task 2).
This module performs the detailed performance evaluation of the packet radio deployments
and requires the following input:
- Terminal location, traffic, and mobility files from the Scenario Generator.
- Repeater location and routing procedures from the Configuration and Routing
Module.
- PR operational parameters supplied by the user; these parameters are discussed
in Section 2.5.2.
In addition, PRSIM accesses a Link Model file in order to determine the network
connectivity.
Given the above information as network configuration, traffic, and protocols, PRSIM
simulates the network operation to measure accurately such quantities as:
2.4
nac_- Average one-way packet and message delay (excluding ETE ACK)
- Average round-trip packet and message delay (including ETE ACK)
- Buffer utilization
- Throughput.
These statistics are more precisely defined in Section 3.2.5.2.
D. Communications Link Module
The Communications Link module is used to determine whether the two PR units can
communicate directly without the aid of any intermediate repeating devices. If they can, a
link between the PR units exists.
Two criteria were employed to define the links in the scenario networks, namely, a
line of sight criteria and an attenuation loss criteria, based on the Longley Rice model.
MITRE Corporation supplied the terrain profiles for the LOS criteria. In this case, it was
assumed that devices could communicate if an only if they were in LOS of each other. In all
cases, the free space attenuation loss was less than 150 dB, the threshold attenuation
specified by CORADCOM.
Alternatively, using the Longley Rice program supplied by CORADCOM as a
subroutine, the Link module can compute the transmission loss between any two PR units.
Then, if the transmission loss does not exceed a specified threshold (which is a function of
data rate), it is assumed a link exists between the two units. The transmission loss
computed by the Longley Rice for two PRU's, A and B, is independent of which of the two
PRU's is transmitting or receiving. Thus, if A can hear B, then B can hear A, and vice versa.
In a graph theoretic sense, the PR network is said to be undirected. The analogous
ssumption was made for the LOS model.
In order to account for the effects of topography on transmission loss, the Longley
Rice program employs an interdecile range parameter .Ah. The computation of Ah requires
a digitized terrain data base, but the Longley Rice program itself does not require such a
data base. The values of Ah to be used in the military scenario have been computed for the
Fuld Gap region of the Federal Republic of Germany by Electromagnetic Compatability
2.5
0
Analysis Center (ECAC). As shown in Figure 2.1, the Link module accesses this Ah data
base. Also as shown in Figure 2.1, the Link Module is accessed by the simulation,
connectivity, and configuration and routing modules. In both eases, PRSIM read the network
links from a file.
E. Connectivity Module
The Connectivity module was used to perform a preliminary connectivity analysis on
the terminals defined for each subnetwork and deployment. This module identifies maximal
independent subsets of terminals, i.e., subsets of maximal cardinality such that for any PR
in a subset there exists a link between that PR and some other PR in the same subset, but no
such link exists between that PR and a PR not in the subset. In order to determine
connectivity matrices, the connectivity module accesses the data base module for PR
locations and the link module to determine which pairs of PRS can communicate directly.
This module uses a depth first search algorithm, which is described in the Appendix.
This module was extensively employed to determine the connected subsets for
networks whose links were determined using the Longley Rice model. The results are
described in the Appendix.
F. Configuration and Routing Modules
In the military scenario prepared by MITRE, terminal locations and needlines are
specified, but the network configuration and the routing procedures to be used in the PR
subnetworks are not defined. The network configuration problem consists basically of three
parts:
- Repeater location
- A PRU transmission power specification
- Transmission data rate specification.
The repeater location problem Itself consists of the following two subproblems:
2.6
Identification of terminals (defined in the C2 , Field Artillery, and ADA
subnetworks) that should also function as repeaters.
Identification of locations where dedicated repeaters should be placed.
It is assumed that all terminals have the integral relay capability and that there are no
additional constraints (in terms of cost or performance) imposed by including this capability.
The number of repeaters required is primarily a function of the survivability criteria,
as well as the PR network delay-throughput requirements. The survivability criteria require
that at least k node-disjoint paths exist between any two terminals that must communicate;
k is a parameter to be determined. The repeater location problem must be solved in
conjunction with a definition of the routing procedures within the network. Additional
repeaters may be required to provide additional communication paths in order to eliminate
congestion. The point-to-point routing procedure defined in the test bed Channel Access
Program 4.6 was employed. Hence, the routing problem consists of:
- Defining primary communications paths through the network for each needline.
- Defining good neighbors to be used in alternate routing.
The other parameters of the network configuration problem PR transmission power
level and transmission data rate are also intimately related to the routing problem. The
power levels and data rates currently used in the DARPA test bed were used in this study.Both terminals and dedicated repeaters operate at 100 Kbps even when they are functioning
as repeaters. The appropriate power level is included in the threshold values specified by
CORADCOM for determining connectivity.
Because of the substantial connectivity in the PR networks studied, dedicated
repeaters were not required in the simulation. Also the projected traffic loads were very
light, hence shortest path routing was found to be acceptable. The shortest path routing
procedures were included in the scenario generator. For the host configuration with linksdetermined by the Longley Rice model, a special procedure was devised to select a shortest
path that minimizes congestion. This special procedure is described in Section 6.
Because of the simplicities introduced by using LOS links, the configuration, routingand scenario generation programs were combined into a single module. Thus, the combined
2.7
0
output of the Scenario Generator and Routing modules consisted of the traffic profile with
the assigned route for each particular message.
2.2. Data Base
The Data Base module consists of three files which contain information exclusively
derived from values provided by MITRE in their recommended military scenario. The only
preprocessing that will be done is a minimal amount of algebraic transformations done for
convenience. The information supplied consisted of:
The three Command and Control (C2), Field Artillery, and Air Defense (ADA)
traffic matrices.
Nodal location coordinates, initially, along Defense Line ALPHA, and along
Defense Line BRAVO.
- Tracks for mobile devices.
The data base module is comprised of three parts subsequently referred to as:
- Terminal location data base.
- Traffic data base.
- Mobility data base.
These data bases are described below.
2.2.1 Terminal Location Data Base
The Terminal Location Data Base, stored on a data file, contains the location of each
PR comprising a scenario. The accessing module reads the file and identifies the location of
a PR unit for the particular scenario being simulated. The Terminal Location Data Base file
contains the following information for each packet radio unit:
2.8
nac_Subnet indicator
Identifying packet radio number
Identifying acronym
Identifying description
- Antenna height
Location (X-Y coordinates) for each applicable deployment of the three
specified. Some units are lost, captured, or destroyed and are thus included only
in one or two deployments.
For the Command and Control, Field Artillery, and Air Defense subnetworks, the
above information is included in Tables 2.1, 2.2, and 2.3, respectively.
The subnetwork identifier is a 3-character binary variable where:
- The first character is I if the corresponding PRU is in the C2 subnetwork; zero
otherwise.
- The second character is 1 if the corresponding PRU is in the Field Artillery
subnetwork; zero otherwise.
- The third character is I if the corresponding PRU is in the ADA subnetwork;
zero otherwise.
For example, if the PRU identifier is 110, the PRU is in the C2 and Field Artillery
subnetworks, but not in the ADA subnetwork. The identifying PRU number is a unique
number associated with PRU and corresponds to Table I in the MITRE report R 1.thSpecifically, PRU J in the data base is the j entry in Table I. The identifying acronym will
be used by the other software modules in identifying PRU's, while the identifying
description provides more detailed information to be used in relating the unit to the
terminology employed In the MITRE report. The antenna height is included for use in the
Longley Rice transmision model; CORADCOM specified the following antenna heights:
2.9
nlac.
0 - ON 4 -, M ,O O
N wtWO O N tVW(V NQ)0
0 ~ ~ j) In 4 ~~~ 4 M'CO- fl 0 0 -40 0~ 0'. 4 M. In -- (1 0v mn M~ 94 0.
0o--o()V )N' ,m4oc )()' )-o'0vP 'oIn'o n)n 0n(C' Vn)nqt N)In n OD4 M
zz0* 4ZNVD00 t% 1 %0 NM00D4 0 1 tR 0
E_-ieoOk N i ON NO V)VZNN 0O 0 Z-N
4.~~m N~CI' OZOZ Z a'.'-C~~~ Z wmm W4 NO. Cl- "~(U ~U
.- wwwwwwwwww --.- *z 4.c.-wwww.a>zw
QQ Q QQ2o z z z w0- 0 IiI_? zI I ZZ OZ 0-~ IO ( 'Oa
0 c (J 4rzzz u4oJ 0 OOI-0X0( C 0~ee.aC L I I IZ0wW4 CieZZZZZZZZC) -. UUZ OU Dzomzteaeio
0000 ~ eizo~000-JW~ILA ZZZZZZZZZI>W4-._ ez)=OcflWO)O)Ozz
>"-o-4zzzz>'.Or% z~mxw""Zw w 4 < 44 4< < 4 > ZW -- 4 D - -- -- W-WWW
00zzz8OOO0zooQ
Wm .104
lIS PAGE IS BEST QUALIT
TABLE 2.2: FIELD ARTILLERY MATRIX PRU LOCATIONS
DeploymentAntenna
ID No. Acronym Matrix Entry Height Initial DL Alpha DL Bravo
1.0 1 DI.V MRIN DIVI-SION FSE B.O 359 269 297 a68 -31- 24611.0 2 1BDE. BRI.GRDE 1 2..O 552 398 504 412 409 4201-1-0 3 aDDE BRI.GADE a 2.0 545 274 458 267 307 2551-0 4 3BDE BRIGADE 3 a.o 528 142 41.4 133 302 1.17110 26 DR DIVISION ARTILLERY 2.-0 391t 308 306 254 1.53 26001.0 a7 TAB TARGET ACQUSTON BAT 2.-0 405 245 342 246 184 27001.0 8 DA-SURV 1. DR SURVEY SEC 1. 2.0 401 a95 313 280 1.77 298010 29 DR-SURY 2 DR SURVEY SEC 8 2.0 400 a45 334 a36 186 268010 30 SIF cC SIF CONTROL CENTER 2..0 631. 392 558 387 47.7 364010 31 SF cC 1 SI.. ONTROL CENTER a.o 608 175 501 161010 32 TP125.58 GROUND SURVEY RADAR 2..0 659 209 548 273010 33 TP036-1 COUNTER MORTAR 1. 2.-0 635 418 552 399 452 393010 34 TP036-2 COUNTER MORTAR 2 2.0 621 a86 523 260 373 267010 35 TP036-3 COUNTER MORTAR 3 2..0 578 1-44 486 1-46 352 1.1701.0 36 TP037-1. COUNTER BATTERY 1- .0 588 373 486 389 424 427010 37 TPQ37-2 COUNTER BATTERY 2 2.-0 533 177 41.2 163 317 132010 38 1--,GMD-1. GROUND MET DET 1. ..0 662 359 561 376 475 380010 39 GMD-,2 GROUND MET BET 2 2.-0 598 608 496 175010 40 RPY $CS BN BN RENT PlL VEH GCS1 2.-0 555 386 520 414 4.06 417010 41 RPV-'GCSBN2 BN RENT IL YEM GC02 2.0 544 283 467 27 317 252010 42 RPV-GCSBN3 BH REIT PI.L YEN GCS3 2..0 533 1.35 423 133 308 1.1.4010 43 RPV-,GCS DR DR REIT P-L YEN GC1 2..0 396 311. 252 317 173 25401-0 4<. RV-GCSFSE FSE REMT P.L VEN 1 a.o 378 272 288 262 149 232010 45 DR SOTAS DR SOTAS 2.0 392 303 295 850 155 266010 46 FSE SOTA FSE SOTAS a.o 362 a65 296 275 138 246010 47 1-40 SURVI B1 SURY 1 1-40 2..0 605 429 531 4a9 451 409010 48 1-40 SURY2 B1 SURY 2 1--40 26.0 592 392 58 380 436 379010 49 1-41 SURV1 BN2 SURY 1 1-41 2.0 598 317 521 299 356 276010 50 1-41 SURVa BNa SURY 2 1-41 2..0 57a 251 498 254 337 234010 51. 1-42 SURV1. BN3 SURY 1 1-41- a.o 532 177 463 158 359 139010 52 1-42 SURa BN3 SURY 2 1-41 a..o 565 135 456 113 340 092010 53 F AA1-401 FORWARD CBS 1 1..5 684 466 591 475 533 460010 54 F RA1-40a FORWARD OBS 2 1.5 682 452 591 455 538 423010 55 F0 A1-403 FORWARD OBS 3 1.5 660 450 584 438 521 406010 56 F0 B 1-401 FORWARD CBS 4 ,..5 660 424 583 426 513 383010 57 F B 1-402 -FORWARD CBS 5 1.5 668 417 592 420 485 351010 58 FO B 1-403 FORWARD UBS 6 1.5 678 408 597 403 463 329010 59 F C 1-401 FORWARD OBS 7 1..5. 693 383 598 396 454 305010 60 FO C 1-402 FORWARD O3$ 8 1.5 685 372 59a 578010 61 6O C 1-403 FORWARD CBS 9 1.5 693 355 594 353 433 287010 62 F0 A t-.4t FORWARD CBS 10 1.5 672 330 593 330 409 847010 63 PR0 A-418 FORWARD BS 11 1...5 686 326 588 31.7 405 22010 64 FO A 1-413 FORWARD S 12 1.5 669 298 591 297 407 191010 65 6023 1-411 FORWARD CBS 13 1.5 664 as5 583 81 417 16401.0 66 6 3 1".41- FORWFAI)D BS t4 1..5 665 874 57.6 866 418 148010 67 60 1-413 FORWARD CBS 15 t.5 668 864 571 256010 68 6O C 1-411 FORWARD CBS 16 15 654 850010 69 F C t-41 FORWARD CBS 1? 1.5 656 830 544 8351O0 70 60 C 1-413 FORWARD CBS 18 1.5 661 814 558 82?
010 71 F A 1--48 FORWARD IDS 19 1.5 644 198 540 803
0t
NI AEIS BSi&T QUALITX £I.D lacmA"I~i CG)X YL&Xii&D TV !)W.
TABLE 2.2: FIELD ARTILLERY MATRIX PRU LOCATIONS (CONTINUED)
DeploymentAntenna
ID No. Acronym Matrix Entry Height Initial DL Alpha DL Bravo
01.0 ,72 RD A .-429 FbRWRRD DES 20 1-.-5 630 177 536 t8501-0 - 73' FD A 1-423 FORWARD DBS 21- 1..5 628 160 533 17601-0 74 O B 1-42 FORWARD OBS 22 1.-5 629 151 529 14801.0 75 FO B 1-422 FORWARD DPS 23 1,-5 622 133 532 13901.0 76 FD P t-423 FORWARD DBS 24 1..5 624 t19010 77 FO C t-421- FORWARD OBS 25 1..5 64a 098 528 118 421 137010 78 FO C 1-422 FORWARD OBS 26 1..5 629 088 5t9 10 4t2 11-40.0 79 FO C -423 FORWARD OBS 27 1.-5 635 075 503 076 4t4 0930.0 80 RD B 1-40 3Ni. REl-L OSERVA t.-5 620 42O 586 41.0010 81t RD B4 t-4.t B2 AERIAL OBSEPVR t.-5 620 280 560 270010 82 AD B t-42 BH3 RERIRL OBSERVA 1.-5 580 1-50 500 090t1O 83 DS B 1-40 DIRECT SUPPORT B1 2.-0 58t 415 51- 408 426 407110 84 DS B 1--41 DIRECT SUPPORT 3h2 2..0 576 275 480 269 327 246110 85 DS B 1-42 DIRECT SUPPORT B3 2..0 547 t47 445 150 326 t13110 86 GS P14 t-43 GENERAL SUPPORT BN 2..0 452 242 356 246 164 268ttO 87 FR GP FIELD ART LERV G 2..0 362 324 307 279 1-44 269110 88 DSR2-63 DS REI-FORC 3h t 2.-0 576 425 504 394 429 431-110 89 DSR2-637 DS REI-FORC B 2 2..0 575 302 499 288 33t 266110 90 DSR2-638 DS REI DRC B 2..0 541 161 438 115 331 135t10 9 6SR2-61.8 DS GEN SUPP B 2..0 472 278 358 269 181 295010 92 BCS-40 A DS BAT COM SYS 311 2..0 618 448 527 437 449 41-010 93 BCSt-40 B DS BAT COM SYS B1 2..0 627 426 530 374 441 39:3010 94 BC$-40 C DS BAT CON SYS B41 2.0 626 3970W0 95 ECSI--41. A DS BAT CON SYS BH2 2.-0 612 315 522 2% 351 248010 % ECS1-4t B DS BAT COM SYS BH2 2.0 606 275 5t 267 336 205010 97 BCS$1-4 C DS BAT CON SYS BH 2.0 60 242 493 235010 98 BC$1-42 A DS BRT COM SYS B3 2.0 577 176 469 178010 99 SCS1-42 B DS BAT COM SYS B3 2..0 564 148 466 158 346 11.4010 $00 EBCSI-42 C DS BAT COM SYS BH3 2..0 567 116 453 123 349 09401-0 -0- BCS1.-43 A GS BAT CON SYS BN 2..0 492 266 390 245 830 299010 1O2 BCSI-43 B GS BAT COM SYS Bh 2..0 475 233 369 2a6 a11 258050 t03 BCSt-43 C 65 BAT COM SYS B 2..0 475 208 349 t99 206 1950-0 t04 BCS2-63A RDS BAT COM SYS B1 2..0 61-7 465 524 427 451- 458010 t05 BCSB-63- RDS BAT CON SYS BH 2.0 618 434 535 404 449 4350-0 106 SCS2-,63C RDS BAT CON SYS 311 2.0 617 403 526 383 449 41.1.010 107 BCS2-637R RDS BAT COM SYS B1 2.0 621 332 517 315OO t08 BCSB-,6372 RDS BAT CON SYS B12 2.0 607 293 51- 278 410 312OO t09 SC-637C RDS BAT COM SYS B12 8.0 579 237 499 252 375 278OtO tO B$CS-638A ADS BAT COM SYS B3 8..0 561 1-98 455 38 360 163010 t1- BCSa2-638B ADS BAT COM SYS BN3 2.0 569 166 457 t04 365 41.010 112 SCSa-638C ADS BAT CON SYS B3 2.-0 560 87 437 079 364 125010 1t3 BCS-6t8A RGS BAT COM SYS PGSt 2.0 498 385 4t6 3t8 a54 332010 114 BCSO-6t8B AGS BAT COMI SYS RGSt 8..0 480 313 4t9 899 812 876010 t5 $CSa-,6t8C RGS BAT COM SYS RASI 8.0 495 a84 399 272 806 835
2.12
NfIS PAGE IS 61f U.lri±LL.iM
TABLE 2.3: ADA MATRIX PRU LOCATIONS
Antenna DeploymentID No. Acronym Matrix Entry Height Initial DL Alpha DL Bravo
1.01. t1.6 .-441. CVB C/ BN 2.0 397a a78 353 284 117. 238001. 1.1.7 A:IC BTRY C/.V BATTERY 2.-0 365 a69 286 272 10 247001. 1.18 B/.CV BTRY C.V BATTERY 2.0 459 a39 378 272 161. 272001. 1.9 C/I.CV TRY C/V BATTERY 2.0 525 t38 464 263 323 249001 1.20 D/.-C9 BTRY C/V BRTTERY 2.0 557 a73 420 132 330 127001. 121. ./R-.C PLr C/V PLATOON 1 2.0 371 260 31.0 278 092 262001. 122 /A/. PLN C/V PLATOON 2 2-.0 355 289 278 258 1.27 232001. 1 33 xV PLN C/V PLATOON 3 2.0 367 269 293 273 1.31. 251.001 t24 /B/C PLN C/V PLATOON 4 2..0 496 32 403 275 a35 287001. 125 B/B/CVPL C V PLATOON 5 2.-0 481. 209 41.8 320 206 233001. 1.26 3/BC9PLN C/V PLATOON 6 2.-0 469 a80 355 243 1.78 29500 127 1/C/CPLN C/V PLATOON 7 a.-0 575 t.79 492 287 329 26300t 128 2/CACVPLN C/V. PLRTOON 8 2..0 530 t42 477 269 324 239001 1.29 3/C/VPLN C/V PLATOON 9 2.-0 565 11.7 452 265 303 251.001 130 1/D.CVPLM C/V PLATOON to 2.0 573 299 456 1.43 328 1.32001. t31. 2/.DCVPLN C/V PLRTOON 1.1 2..0 548 274 444 1.13 328 1.1.100. 1.32 3/D/VPLN C/V PLATOON 1 2.-0 578 240 413 1.39 301. t14001 133 ICFV/.A FIRING UNIT PH 1. ..0 362 306 322 307 25 31.6001 134 2CFVAR FIRING UNIT PH 1 8..0 345 294 302 301 21.7 308001 1.35 3CFV/R FIRING UNIT PH 1 ..0 375 286 338 289 238 29200t 1.36 4CFV/R FIRING UNIT PH 1. .-0 337 28t 281. 285 198 273001. 137 5CFV/A FIRING UNIT PH 2..0 395 270 385 262 a37 262001 138 6CFV/A FIRING UNIT PH 2 2.0 345 268 265 262 1.95 249001. 139 7CFV/R FIRI.NG UNIT PH 2 2.0 374 266 312 245 231. 21.8001t 1.40 SCFV/R FIRIMG UNIT PH 2 2..0 365 260 27 237 193 198001. 141. 9VFV/.A FIRI.NG UNIT PH 3 2..0 358 274 298 272 1.81. 299001. 1.42 $oVFV/R FIRING UNIT PH 3 2.-0 362 71. 301 269 t79 293001 1.43 t1VFV/AR FIRING UNIT PH 3 2..0 356 269 292 268 t65 273001. 144 1.:2VF/A FIRING UNIT PH 3 .-0 359 267 297 265 165 262001. t45 t3CFV/.B FIRING UNIT PH 4 .-0 521. 329 423 328 115 280001. t46 14CFV/B FI.RI.NG UNIT PH 4 2.-0 529 311 446 310 080 278001. 147 tSCFV/B FIRItNG UNIT PH 4 2..0 507 299 436 292 1.29 268001 t48 1.6CFV/B FIRING UNIT PH 4 2..0 521 275 409 268 076 26500t 1.49 1.7CFV/B FI.RI.mG UNIT PH 5 ..0 51.3 a48 409 247 1.55 26100t t50 1.CFV/B FIRING UNIT PH 5 a.-0 498 227 400 21.9 085 251O 1.51. 1.gcFV-'B FIAING UNIT PH 5 2.0 505 206 392 201. 101. 23300t t5a 20 CFv/B FIRING UNIT PH 5 a.O 50a 177 379 1.83 3 29001. 1.53 a1FCV/B FI.RING UNIT PH 6 2.-0 47-2 285 361. 271 32 248001. 54 a2VFv/B FIRING UNIT PH 6 a..0 475 282 359 268 1.34 245001 1.55 23VF/ FIING UNIT PH 6 8..0 455 244 358 246 1.29 244001. 1.56 84YFV/B FI.tI.NG UNIT PH 6 .0 454 242 360 244 1.31. 242001. 157 25CFv/C FIRING UNIT PH 7 8..0 602 1.93 501. 291, 334 26800t 158 86CFVA.C FtIING UNIT PH 7. a..o 607 178 498 286 331 26400t 1.59 87F%'.C FIRING UNIT PH7 . 8..0 588 1.08 482 273 306 25900FR 1.60 88MVC FIRING UMIT PH 7. .O 576 085 459 272 31.2 255001 61. 89YPYAC FING UNIT PH 8 8.0O 5?.9 179 479 868 303 25500 1.68 30WVVC FIRNGS UNIT PH 8 8.0 579 1.74 46t 265 307 251001. 1.63 31YFV/C FIRING UNIT PH 8 a..o 524 1.46 456 863 329 248001. 164 32PV/C FIRING INIT PH 8 8.0 533 44 46a 871. 3a6 243001 165 33VFV/C FIRN UNIT PH 9 8.0 588 1.40 509 29t 396 304
2.13
TABLE 2.3: ADA MATRIX PRU LOCATIONS (CONT.)
DeploymentAntenna
ID No. Aeronym Matrix Entry Height Initial DL Alpha DL Bravo
001. 166 "34VFV.C. FI.RING UNIT PH 9 2..0 523 133 500 a82 381 875001 167 35VFV-C FIRING UNIT PH 9 ..0 570 -1.7 487 265 373 a38001. 1.68 36YF',AC FIRING UWI.T PH 9 a..O 566 114 474 a55 371 198001. 1.69 37VFV/D FIRI.NG UNIT PH 1.0 2.0 575 303 448 1.52 331- 1-37001. 1.70 38VF/D FIRING UNIT PH 10 2.-0 575 298 443 148 333 130001. 1-71- 39VFVAD FIRING UNIT PH 10 2..0 552 a81 412 137 304 1-1.9001. 1.72 40VFV/D FIRING UNIT PH 10 a.o 547 277 41.5 136 308 117001 173 41.VFV.D FIRING UNIT PH -1. ..0 546 a70 41.1. 1.31 328 116001. 1.74 42YFVAD FI-RIRIG UNI.T PH -1. 2.0 55a a69 41.? 131 a96 115001. 175 43VFVAD FIRIG UNIT PH 1.1 2.O 58a a40 442 11.9 323 1.1.1001. 176 44VFV-.D FIRING UNl.T PH 11 a..O 582 235 439 1-1- 301t 1.1.1001. 1.77 45CVAD FIRING UNIT PH 1- a.-O 586 31-8 469 1-60 376 1-57001. 1.78 46CFV/D FIRING UNIT PH 1.a a.o 53 a89 469 142 366 1-38001. 1.79 47CFV/D FIRING UNIT PH 12 B.O 58a 261. 463 1.1.5 358 -1.8001. 180 48CFV/D FI-RING UNIT PH 12a 2.O 578 224 462 090 349 1.0t2001. 181 1 FAFR FRIWRDFRER L.RTRRDRI 2.-0 647 459 579 462 520 418001- 1-82 2 FAFRI FRWRDARER RLTRIDR. 2.O 661 395 582 386 446 31-1.001 183 3 FFIP FRWRDFRER RL:RTRFRDR3 2.0 628 318 575 325 402 268001 1.84 4 FRRR FRIJRDRRER RLRTRIDR4 2.0 625 268 544 283 391- a01-001. 1.85 5 FR: FRWRDRRER RLRTRRDR5 2.0 638 208 539 242001. 186 6 FRAFR FRRDARER RLRTRRDR6 B.O 617 161 535 1.83001. 187 7 FRAR FRWRDRRER RL-RTRRDR7 2.0 608 127 51-9 137001 18 8 FAFR. FRURDRRER RL.RTRRDR8 .0 628 089 502 089 388 1-04001 189 1-451 IH MPROVED HRdK TSQ73 2.0 408 287 305 316 1.27 e1-4001 1.90 R/I.H BTRY IMPROVED H:WK BTRYI. 2.0 515 434 415 433 276 417001 191 S/AIN ITRY It2 a..o 422 328 398 301 228 1-99001 1.92 CxlH BTRY 3 2.-O 47-7 1% 354 166 205 11-3001 193 1.FU/I-451 "- FU a..o 5O 439 424 428 292 424001 194 BFU/.-.451 " " a 2.0 508 436 423 436 278 409001 195 3FU/1-451 3 2.-0 51.4 428 423 419 a89 403001- 1.96 4FUA.1-451. 4 ..0 429 339 397 31.1 241 206001. 1.97 5FUAI-.451 5 2.0 425 335 392 303 239 203001. 1.98 6FU/1-45 6 2.0 428 331 390 299 235 195001. 199 7FU/-1-451 7 .0 481 aO2 357 1.94 19 1.31001 200 8FUAI-451. " 8 .0 467 190 354 1.88 213 1.8001 201 9FUA1-451 " ". 9 2.0 47-5 183 355 1.79 2O8 106001 202 1RE/1-77 REDEYE I 1.0 629 458 587 473 535 423001. 203 2REA.--77 " 2 1.0 647 443 57a 461 504 419001 a04 3REA1-77 It 3 1..0 644 427 548 459 511 41.7001 205 4REA.-.77 I 4 .o 642 41.7 584 457 522 409001 206 5RE1-77 Ot 5 1.0 600 458 582 437 507 396001 a07 6Ei1-n at 6 1.0 645 404 581. 419 417 128001 208 7REA1-.7 " 7. 1..0 65a 397 568 414 401t 11.9001 209 8REi-78 " 8 t. 667 379 569 405 397 113001 210 9RE1-?8 ". 9 t.O 635 394 595 404 404 107001 211 RGA1-.8 l o 0 I,O -699 47-9 699 47.9 404 210OOt 2a ii E-9$-.8 "t It 1.0 677 3W7-001 213 12REA.7-9 "t 12 1.O 67a 31.8 559 334 470 335001 214 -3RExt-.79 "t 13 1..O 659 339 566 326 456 326001 215 t4RES/-79 $ 14 1.0 664 309 585 320 449 318
0 2.14
THIS PAGE IS BEST QUALITY PRACZLC$LBli cftOM "Yt YL~FPlib&Da IOVD
TABLE 2.3: ADA MATRIX PRU LOCATIONS (CONT.)
Deployment
D NAntennaID No. Acronym Matrix Entry Height Initial DL Alpha DL Bravo
001 216 15RE'1-79 1.5 1..0 659 280 566 a55001 217 16RA1--.80 16' 2.0 641 255 579 278 384 270001 218 17RE..- 0 17 I-O 633 228 550 224001 219 18RE1-80 " 1-8 1.0 612 243 564 a75 394 265001 2aO 9RE.1-80 1-9 1.0 631- 250 546 282 403 243001 221. aORE4I-O " 0 1.0 640 246 575 270 401 226001 222 alRE--80 1 1.0 675 368 589 373 485 376001- 223 2RE1--81- a l.O 669 354 554 365 507 373001 224 83RE1-81 3 1.0 657 363 572 365 488 372001 825 24RE/-.81 4 1.0 669 406 585 365 485 353001 226 25RE/1-1 5 1.0 678 379001 227 26RE/1 3 a6 1.0 680 454 379 2 06001 228 7K/.E1-3 7 1.0 .371 238 402 185001 229 8R E-1-23 28 1'-0 365 247 407 174001 230 29R6/1-23 " 29 1-.-0 629 470 501 476001- 231 30RE.1--23 " 30 1-.-0 594 475 527 452001- 232 31RE/1-82 31. 1.0 620 146 501 075 414 085001 233 32RE.1-8 32 1.0 629 165 532 173001 234 33RE.1-8 33 1.0 602 141 501 100 405 076001 235 34RE1-82 I 34 1.0 595 132 493 098 393 073001 236 35RE/-82 " 35 1.0 608 128 513 097 423 072001 237 36RE/- 36 1.0 613 071 525 148001 238 37RE/1-2 37 1.-0 601 079 508 145 407 160001 239 38RE-1-2 " 38 1.0 614 079 502 135 393 1-60001 240 39RE-1-2 39 1.0 626 074 524 133 384 156001 241 40R51-2 40 1.0 374 219 406 a62001 242 41RE-1-3 41 1.,0 669 338 585 304 455 312001. 243 42RE/-3 - 4a 1.0 663 294 552 251001 244 43REA1-3 " 43 1.,0 630 289 558 249001 245 44RE-1-3 " 44 1."0 657 266 542 245001 246 45R1E.-3 45 1.0 646 a75 558 a38001. 247 46RE/1-4 " 46 1.,0 646 201 51-9 120 395 067001 248 47R-E--4 47 1..0 626 19 538 204001 249 48RE-14 " 48 1.0 609 188 512 195001 250 49RE/1-4 " 49 1.0 629 178 519 189 '001 251 50RE1-4 " 50 1.0 598 107 536 189001 252 51RE.1-5 " 51. 1.0 613 245 585 888 41C 283001 253 52RE.1-5 " 52 1.0 637 237 555 a30001 254 53RE.1--5 53 1.0 6a5 aao 519 aa4 J001 255 54RE1-5 54 1.0 636 223 540 222 '"001 56 55RE"-5 55 1.0 639 216 549 2001 56R51-40 56 1-O 623 448 530 438 48t,36a001 258 57REt1-40 57 1.0 631 488 51-3 409 448. 393001 a59 5RE-.40 58 1.0. 589 41a 538 397. 44.371001 860 59E'.1-40 59 1.,.0 632 398 1.001 261 60.1---41 60 1.0 616 314 530 299 33a 5.001 268 61tRE/1-41 - 61. 1.0 580 277 495 271 358 a47001 863 68RE1--41 62 1..0 613 275 518 267 34a '4.97001 264 63RE.1-41 63 1-.0 606 a39 497 234001 65 64E'1-,48 64 1.0 579 176 475 175 MC...
Lis
naci
TABLE 2.3: ADA MATRIX PRU LOCATIONS (CONT.)
DeploymentAntennaID No. Acronym Matrix Entry Height Initial DL Alpha DL Bravo
001 266 65RE-1-42 " 65 1.0 569 1I49 46 157 354" . t5'001. a67 66RE'.-4a - 66 t.o 549 47 447. 148 330'. -144001. 268 67RE/.t-42 67 1.0 573 t14 455 a7 344':,!09001 269 68RE/1-43 . 68 1.0 496 a67 359 1.99 P35 .',699001. 270 69RE1.-,43 69 1-.-0 455' 24a 358 246 t60-t6"9001. 271- 7.0RE'1-143 70. 1.0 402 a34 395 242 21.4-,,158001 27a 71.REA.-43 71. 1.0 482 a09 375 8a6 214,19400t 273 72RA 2-.631. 72 1.0 622 464 526 4a7 4.53, ".4*q001 274 73REA2-.631- 73 .0 6aa 430 536 405 4524 *3
- 001 275 74REAB-631 74 1..0 58t 4a5 507 396 431.',<433001 276 75RE'2-631 75 1..0 625 404 532 383 457., 4.3001 277 76RE-a-637 76 1.0 626 330 524 314001 278 77REA2-637 77 1.0 578 302 505 287 41...000t 279 78RE2--637 " 78 1.0 609 a93 523 a77 37'wa77001 280 79REAB,637 79 1.0 583 a38 506 250 333'i363001 281 8ORE.2-638 80 1.0 566 197 455 127 364.163001 a82 81RE/2-638 1 1.0 574 169 422 15 367.;39001 283 lRE/2-.638 " 82 1.0 543 162 464 t03 335 * 13600t 84 e3REa-638 .83 1..0 563 124 445 078 36-4 4,'2Q400t a85 "4REB-6tS 84 1.o 504 330 4a1 322 a591b43a900t 286 85REA2-6 8 95 1.0 489 314 482 297 18rq500t 287 86REB26t8 96 t.0 500 a85 40t 272 . t 4o 8001 288 87REA2-.618 87 t.0 475 281 362 a69 a*001 289 88RE.1.--451. 88 ,.0 521 434 4a6 436 27V-9. 14 8001 290 89RE1.'451 89 1.0 432 33t 399 307 23k1,,aie3001 291 90RE-t-451 ". 90 t.O 409 a89 357 165 a@:o.4S
L 001. 29 9tRE/1-451. 91 1..0 482 95311 314 1.3B7Q5OOt 293 9aRE$1-441 9a "1.0 405 276 466 2-57 .a1 _V1 45001 294 93RE1,-44 " 93 " .0 369 270 4a5 31- 1ai.,36001 295 94R1E.-441 94 ..0 465 238 384 270 165,1127900t 296 95RE/1-441 95 1.0 534 138 288 274. 3aa744001 297 96RE/.1-441 96 .0 562 2?7 355 388 33'4 134
THIS PAGE IS BEST QUJALITY ME1'.I~CABr S ,"..
2.16,
REDEYE, 1.5 meters
Forward observers, 1.0 meters
All other units, 2 meters.
The coordinates included in the- Data Base are tenths of kilometers from the origin
assigned by MITRE. For example, in initial deployment, the 52ND Mechanized Division
Support Commano (DISCOM) is located at NB325 245. The X and Y coordinates of the PR
unit (PRU) is, in this case, 32.5 km and 24.5 kin, respectively. Let (X1, Y1) and (X2, Y2) be
the coordinates of PRU's N1 and N2, respectively. Then to find the linear distance in
kilometers between Ni and N2, the following equation is used:
distance between Ni and N2 in km= 10 ) 10 "
Attrition occurs twice during the 30-hour period. If a PR unit is lost, destroyed, or missing
as a result ol-attrition, its subsequent coordinates are omitted from the Data Base. Also,
REDEYE units 26 through 30 are held in reserve during the DL ALPHA hostilities, but return
to action for the DL BRAVO deployment. Hence, the coordinates of these units are not
included for the DL ALPHA deployment. Thus, the module accessing the location file for
any scenario will recognize the blank and consider the Packet Radio unit as not existing in
that particular scenario.
2.2.2 Traffic Data Base
The traffic data base contains the traffic data corresponding to each of the three
traffic matrices, Command and Control, Field Artillery, and Air Defense, supplied by
MITRE.
The MITRE traffic information includes:
- Total number of transmissions during the 30-hour scenario
2.17
0
Total bits transferred
Average message lengths
- Busy hour percentage (based on a 24-hour period).
The number of messages transmitted between any two packet radio unit pairs is
independent of deployment. Hence, if there are X transmissions per second between packet
radio unit A and packet B in the initial deployment, then there are also X transmissions per
second between these packet radio units in the Defense Line ALPHA and BRAVO de-
ployments, providing neither unit is lost through attrition. Hence, the overall magnitude of
the traffic flows are changed between deployments only due to attrition, not due to any
change in requirements. However, due to changes in the routing, some devices could
experience an increased loading.
In the traffic matrices supplied by MITRE, each transmitting source and receiving
destination are identified by name, e.g., DIV MAIN CP, BDE (3), FO (9). In the case where
there are no parentheses following the term, there is only one entity; for example, DIV
MAIN CP means that Division Main Command Post is a single entity implying only one PRU
(for a particular subnetwork). Each source may comprise a group of PR units, as indicated
by the number in parentheses (BDE BN (3), FO (9)). For BDE (3) there are three separate
brigades implying three packet radio units, one for each brigade. However for FO (9), there
are 9 groups (corresponding to the 9 Battery Computer Systems units) of Forward Observers,
each composed of 3 Forward Observers; hence, for FO (9), 27 packet radio units are
required.
The functional needlines are depicted in Figures 2.2 through 2.7 for each subnetwork.
In Figure 2.2 the complete C2 subnetwork is depicted with a box associated with each PRU.
The information in the box includes the military name or unit number; the number encircled
near the box is the unique identifying number from the location data base. A line with an
arrow from box A to box B indicates unit A transmits to unit B; in the C2 subnetwork there
are no simplex communications lines, so al lines have arrows on both ends.
The numbers near the arrows indicate the total nuber of transmissions during the 30-hour duration in the direction of the arrow. For the C2 subnetwork the busy hour per-
centage is 10% for all needlines.
2.18
0
_j 0-0L&J 0lC
C.) = LA
im CD -I- DLI2
U.~
C) C.)O
C).
z~ CCV)CDC)C
40% cm > LLi
0 0 0 0i 0 0 0 0D I=0% o 0% 0%. 1- 0.4- 0% 0 0 %
C.)Dz ~ j C Au
0iCD C1
oC
co 004-
LJ c- r40 U)
uJJ 0C
-J LL -i J
0 -. mv LO L.n
X 0-4 L'- 0
-L 0- ::0W-
L)U
'n CDJ cn-
'cr c I-JCo
0 *%j 0.C C©
Q 2- 2.20= - "
cc Ilam
U- w.
Cie La
L-&i C-J CC
a. S-. o o.
cc~ L)IA c0 D-%
Cl.J 0.
5-O
L -i W. - i a
aan
4c-C
Ix I : :- '0-- Ll W=0W w .L
40J cr V c
-C C)C. c
C- LYt 0n
I' ~cc0 0 221 5-____o
La. LIJc
= L:w ) 06P4L- u.m
LLI
enj
Ln -lip-Jl'-4
CD oe
x o0
fnJ
-% U-
M U,
r*- L CD j C>C,~j
uli
0jC.;Q-
R 2.2C)
LL 3LLJ
LrL)
N Lf Lfl~r
I- r- 6-ccenc
LL- RLr) " Lf l a. O
cc~~r CD LA QLki 2C N
z4
I= I
121C U-Liii-
LL- 0. c
LU L U.
0D 40N~n~ Ln n co0i
p La - I---c
coL =C0Ul -
0- )-4AL -
Ln CX r- .l
L*L
c~c coC-
~.- 2.2
w c
0.382xl06 O.382x106
Inn
-4 -. -
ta. 4NC44
-.j -4
c.~ $ ~2 I0a
SSw
~J~® ~XaD
x
w0
244 n qjc
Analogous information is presented for Field Artillery in Figures 2.3 through 2.6.
Multiple figures were employed for clarity of presentation, rather than to define further
subnetworks at this time. The three levels presented are depleted in Figure 2.3, 2.4, and
2.6, but the numerical values of total number of transmissions were not included in
Figure 2.4. Instead, for purposes of clarity, a generic Direct Support Battalion and a single
General Support Batallion are depicted in Figure 2.5. With quantitative values for the
needlines presented in Figure 2.5, the needlines for Figure 2.4 can be readily deduced.
Since the busy hour percentage is a function of the needline, this percentage is
specified in Figures 2.3, 2.5, and 2.6; this is the number in parentheses next to the needline.
In Figure 2.7, the needlines for the entire ADA subnetwork are presented. The major
distinguishing characteristics of this subnetwork are:
- Multidestination traffic
- Traffic rate is uniform throughout the 30-hour period; hence, no busy period
fraction is specified.
In the computation of the needlines shown in the aforementioned figures, the following
convention specified by MITRE has been employed. For single transmitter and receiver the
matrix entry provided by MITRE is the point-to-point traffic flow for the 30-hour period. If
the transmitting entity is a group with no subgroups, e.g., BDE (3), and the destination is a
single unit, then the matrix entry is again the point-to-point traffic. However, if the
transmitting entity is composed of a subgroup composed of three FO's, then the matrix entry
is the traffic flow from each subgroup to the destination. In this case, the point-to-point
traffic flow is computed by dividing by the number of units in a subgroup.
When there are multiple destinations, then the matrix entry is the total number of
transmissions to all destinations. Hence, the point-to-point traffic must be computed by
dividing the number of destinations.
The traffic data base will contain the following information for each scenario:
Total number of transmissions over 30-hour period from each military unit to
each destination unit, i.e., point-to-point traffic.
2.25
Iac
Busy hour traffic percentage, i.e., the percentage of the total 24-hour traffic
that occurs during a busy hour; for ADA this value is 100/24.
Transmitting PRU identifying number.
Destination PRU identifying number.
For DL Alpha and DL BRAVO, 1, if the needline is contained in that deployment,
and 0 otherwise.
This information is listed in Tables 2.4 and 2.5 for the C2 and Field Artillery
subnetworks, respectively.
2.2.3 Mobility Data Base
This part of the data base applies to mobile PR units and constitutes a location data
base for mobile units. It contains:
The coordinates of the points defining the tracks over which mobile PR units fly.
It is assumed that the tracks are in the shape of a quandrangle formed by joining
each of a set of coordinates; this is an approximation to a race track pattern.
- The velocity of motion of each mobile PR unit (meters/second).
Figure 2.8 shows the track for mobile units on the south sector where points PI' P2,
P3, and P4 comprise the set of coordinates defining the track. Each side of the quadrangle
is divided into shorter lengths of sections (as in Figure 2.8), as defined by a sector length
parameter in the data base. This parameter specifies into how many equal length sectors
the track should be divided. Clockwise motion is assumed. The mobility data base is listed
in Table 2.6 with a default velocity of 60 meters/second and sector parameter of 10. The
only scenario with mobile devices is the Field Artillery scenario during the initial
deployment.
* 2.26
TABLE 2.4: COMMAND AND CONTROL TRAFFIC DATA BASE
Dep. IMsgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU DL Alpha DL Bravo
a -0'0 .- - . -
• A .*.j'O .10 -1 * 1 1.1.0oo . 4 .
60.Oi0 "*0 "1* """6i".00 "*' i"* 1 1p0.'00 "*0 "I 0 * "1
0;'00 "A' "* *'"1 -1O'0""0-1- i'O *" --
60.'00 "*0 -* **" *""60:'i,0 -kO *1 *1 1 0* Oo..'00 .10 -* .1': - 060'00 i- -1 14 -* 0?0O.'0 "10 "* *'5 "I "1
• 1"50.'00 "10 "1" "1"6 "I" "*t0".00 "10 "1" "1? "1" 1
• 1bO-.1J1J -0 .* -1-I - I
~0.;00 "10 "1 "k"9. "* "14be.00 .10 .1- 0 -I 1
*kbO:'00 ,10 .1 1, -* I
• l 'a:'0 "10 1e"""I I90".00 "*O b "1"90;'00 "10 ' p ' *"90.'00 "*0 1"
*50"..00 "10 *1 "* .* "*.0.00 "I0 13 "I I
9P0.0O .*0 - -1. 190".00 "10 "I' "1" 0
*50:.00 *.0 4 .* * l90.00 -10 4 *.0 .*--190:'00 -10 4 *.-1- 1- .*9t0.'00 .*0 4 .1. -* 090:.00 "10 4 "I4 "1- 090:'00 "10 5 "I "I I
• 10:00 *.0 b 1. -*90:'00 "10 6 "I "* I
•*io'o'0 "1'0 6 *4 *"
90;'00 .10 -1'*1. .1
t90".00 "10 8 "1" i" "1• .'0".00 "10 8 * 1 "*
90".'00 .10 * 1 -* -1.
'0.'00 i'0 "0 "1 "1"• ' ;O0 10 .10 4 -* .1.
,0;'00 .*0 .*.* .* .I*1• -*..0 ' .1'. 4 -* i
90J.'J0 .*0 .* l i. 0• .h'o io' 1 .1. 0
• e,.',.,0J 1S .1*j."i 0
+ .27
TABLE 2.4: COMMAND AND CONTROL TRAFFIC DATA BASE (CONT.)
Dep. IMsgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU D)L Alpha DL Bavo
* 4 l**0 *4 4u
*0 vi( -1.5 *11.
*5u..*, *1.6
*1-0 *1. e2 +*10 eip +1
85-, *251.*1
8?. ea. *1.Ai'u 85 el *
91.0-0 A-u 26' E? +98?.0 *1. e 0
*u 2i. 05 f1
*aAul *a** e6. *1.90-.-,0 e? 2e:F eeI1u 8 e e
2.2.
nac.
TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE
Dep. I
Msp/30 hr. Busy hr. Trans. Dest.
SPnarjp Pe taJL PU PRU DL Alpha DL Bravo
k9AU -1-0 - ,1 *
S, -100 f 86 -8i03:J0i *5 . * -
50 8 -?-54 50 E 9 49i3
'50 50 82 94 "+5 Ep 84 * *
:5J 50 5 "I' 150 a 95
;?.'50 50 91-*5 4 e5 150) 4 *E 0u5.50 &0 4 950 4 +0A 1-
150'00 *5 26 06 "1'
-:'0 80 8w Jw "1*800".'0 - 0 t -- -i 1 0
' 00:.00 8 86. 80 *"i-15.00 i'0 8w 4 " -
*8 26 8? *
90"'00 *0 e6 3" 1.9:400 -*"0 0w i9 -1 0
•2*-3000 *10 26 5 i i-375.O0 "0 21 6? ' "
-1300.00 *0 8w 83 *"-1• 30000 *0 o 84 i-10;00 *0 26 88 i+.50-'00 *0 86 89 +•150".00 *0 26 90 -
-*50.00 i'5 8w @ 1' -141
8 i5i? 6 +
• " 00-.00 *8, 80 86 -* i
, .*. .00 i8 9*; 88 - 0
• . -00.'00 00 -... 84 4 i" 1..* *O0-,,'O 84 85 -- 5 ,' 4.
" ?" :'00 80 86 8.6 *+1 " *"'1 9.'0 0 *1 8 * "1 *90".t00 i'e9 16 2 61
-80".'00 4,0 88 * i"
S2.29
nac..TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE (CONTINUED)
Dep. 1Msgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU -DLAlpha DL Bravo
.0 46 -1. -1.*0 44 + 1
,1St'j -10 4?] -1*1 1,-'5 45 a6 +1 f1"
•~ 1'0;'J *k5 5* -1 -.,3 4 5".'0 -I0 ,5 O5 - -1-
149 e4+
.'00 - 4 0 -1;3 5-,00 10) 5' -1. .-
54 9
k9&.5;00 -1" 50 ,,.1 1
*5 54 111-9&. -1'5 55 9i +
15 55 9e 1 "•1.'00 *5 56 9a f -1'
• 19v:00 1.5 56 1 +1 I
-1- 5 93 +-15 5b w
1:00 "15 59 94 "1 "1.00 .15 59 e -1 1.
15 6 94 I.00 "15 60 21
;15 6.1. 94 -1 --,.'0 0 5 9 5 *1 *
*5 6- 95 0 0
•'9'O0 "1.5 &" .1•.:00 5 64 95 1
15 64 .3 -1. .1..15 5 96 1.
•96-.00 "1"5 65 1-k.o00 .15 6 " 9.•9"0O "5 66 1
+15 67 96
• 196:'0 "15 o53 "1 i"
• ;00 15 6? - -•.00 "*"5 66 3 - i
*5 69 9a * 0•*98o.'0 '5 89 3 *. 0
*5 60 9? 0 09.'00 *5 10"1" 0
+5 96i'9'.'00 b 1 4 0
'.'00 *5 N' ! * 00 2.30
- -- -- -- -- -- -- --- -- .... .. .. ---
naci.
TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE (CONTINUED)
Dep. 1Msgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU DL Alpha DL Bravo
•9.00 -1-5 :. 4 1. 0+5 96 98
• .. .5 44 9i 10•19u.00 *'5 ? 4.1k.0 *5 99 *1.
-15 ?4 4i'3-0 1 45 ?5 99J
1-96-J.00 4.5 ?5 4 1 u•1.'*.'00 .5 ?6 99 9 9
-1.9 -4.5 40*1*.*0J 5 ?? *0lu
-5 - 4• 1'3:-0 i5 8 4.00 f i
-1' ?8 4 1-1:1 "1'5 ?9 *00 f
•19. 0 -5 79 4 1 .*5 80 0*5 8* 84 "1' 0*5 8a 85 iU*5 Eta ie80 Eta a-
•*100:.00 80 8 *: .i
a8:'00 .10 8i 40•19.00 *5 ea 80 "k u
5500 8a 8? ."i""5.00 15 8a 92 "1 "1*• 1-5:0 0 -1.5 83 9a I .1?5:',0 -1.5 8 94 0 045:'00 -0 83 88 -1 .1• l-u.00 4.. 83 5a - 1• -. 0 *5 a 54•''0 .1*5 8-3 55 "1"•-.0 *5 83 56 i"11900 *5 83 5?
*5 09 58*5 ,V-3 59 i
*9w.'00 .*5 83 60 " 0*5 83 64 * "
30..'00 '5 84 a -143?;'50 4.0 84 * 4• l'88'00 80 04 a4 .
A38.0O 0 84 45 . .* .OO 4, 84 8* 0
150'0 5 8 87 - .*45-.00 *5 84 95 *
-*5:0 -' 84 8 *1 i"i-?5.'00 *5' 84I 9? 4* 0*5:?.00 *10 84, 89 ... i.
*5 i *"
*5 4 *
2.31
Ilaci
TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE (CONTINUED)
Dep. 1Msp/30 hr. Busy hr. Trams. Dest.Scenario Percentage PRU PRU DL Alpha DL Bravo
.50;.00 .10 *. 90 1
45?:0 0 10 0 05*5:'00 5 .1-1
*?-.00 15 90 **0 * +-5:00 -*5 90 .141 .
.- 1-5 90 -1-2f5 9-11 a6
.1S-,O ,5 9k *-5 +
120:00 *5 92 5-3 11.5 4 4 115 9 55 "1"
75;'00 "1-5 92 83 '1-15 93 56"100 1-5 93 5? ,-1*5 93 5E '•.-0 93 40 - 1
•*t- '00 .1'5 9a 83 f-15 94 59 " 0 0
• 1100 65"4 • 0
4-15"A,00 15 94 Oa " 0 00 -5 95 62 +
*5 95 63 " . .1"
*2,:,00 *5 95 e4 ' *1-•1-35.'00 .10 95 4' -1-75..'0 -1'5 95 04 . -
*5 9& 5 "51".00 "*5 96 '6 + A.
.00 1'5 9e l8? 1 0
.35..,00 -10 9 49 4) .1. -1"5:'00 15 98 84 .1 *1"
- -. 00 *5 97 " " 0*. '.O0 *'5 9? 89 .1 (1. 00 51. 9? ' 0
*0 9? 4**0+75-A-0 *5 9 4 * 0
''0 '*5 98 ?e 0-*.-5 00 *0O 98 4 4*'."O0 *5 98 ?3* -.'00 *5 "9 ? 0-. *5 9 05 + -I"
*: *00 *5 99 ?4 ." i* 8~0 *5 99 5 0
*5 '99 760* *85OO *0 99 4e
*5 99 Ob .
* 2.32
iac_TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE (CONTINUED)
Dep. IMsgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU DL Alpha DL Bravo
1.J1.JO .5 U4 o5 - i-1't,-.J00 *15 U4 'A ' A
-1':0 15 U4 6? ' 0
•19.i'0 1'5 U'? 00i 0,U'3?.0 -15 05 4 1" .
-11*50 0 05 1' i"3'0 "10 05 4 ' i-'
lu
1'5 a5 e -1' i
i--.00 5 e5 e?- 1.i
*5 e5 9e '1• -5'00 -I5 5 99 -"-
-1-5 e5 -1-4 3- i-1'57b'00 -J-S 05 9i i.• 196:00 '5 5 "il +I
"1"96'00 "1"5 U5 ?1 e5 ii'965'00 1"S5 ?a 1" 0
• 1"J:'J 15 US ?e. " 01"9t6:00 "1"5 US ?5 "'"'-9-500 e1-5 5 ?6 '0 '
"19t:'00 "15 85 ?9 .1""1•190:50 i'O :?:'?' iI"196:'00 -1-5 e5 7 -i
.1"0:000 .1-0 0 .A'1'0i'5.00 'AS t ' 1" i"
• 1'?5:'00 "1"S u 'A1"01" '1' i"
• 1-85".'0 i'5 8t' "1"0 ."Ai
* :'1'.00 .1" 8? ' 'i" 'i"-1-5:-00 5 8? +S '1' i"
-15 86 89 'A' i"
1'.5:00 5 + 9 i" i• 15:'0 5 e? 9J "1" '"
•150:'00 1"' 0 8 '6 1' '1'451' .0 -A'-0 89 8'i i- '1'•1-5:00 5 e? 08 '1' i"
-1 -a : 0 0 t 5 0 09 1 .0 5 '1 " 1 ''1'5.~0 ''S 88 e6'~
•'150:00 "1) 59 ' '1-451-' 00 1-0 9 4 - '1'
'5.O 5 09 8? " "
*5 09 +0? "•-i5-00 *5 89 *'00 - 1
*5 89 *09 - 'A'
2.33S"
nac.TABLE 2.5: FIELD ARTILLERY TRAFFIC DATA BASE (CONTINUED)
Dep. IMsgs/30 hr. Busy hr. Trans. Dest.Scenario Percentage PRU PRU DL Alpha DL Bravo
re 4;.0) -1-5 .1-00 ?8 1 +.1k:O 15 "100 *g 1" 1
00 -1-4) ?9 ++a5..1'0 14e -1*0*1-
•":500 1"5 "l-2 E, f
•.-o 5 8180 11"75".-00 "1'5 1-05 Ee 1. +
.1" 5;0' "15 l'88e "' .1 "• 1'?5:'040 .1-5 "1'0i E0-9 f1 4
.'5 -08 e9 "1 "..5 .109 E*9
t'?5;00 15 .101 90 -1*1
04) .1-5 11 *-1
1-75;0 -1.5 -1** 9. -1 f
2.34
(590, 190) 8S (630, 190)
S6
S4
S3
(570, 070) (600, 070)S5 S4
FIGURE 2.8: EXEMPLARY AERIAL TRACK ON SOUTH SECTOR
ON 52D MECHANIZED DIVISION
I* 2.35
TABLE 2.6: MOBILITY DATA BASE
Identifying SectorNumber PI P2 P3 P4 Velocity Parameters
79 630 460 660 460 670 340 630 340 60 1080 640 330 670 330 640 210 620 210 60 1081 590 190 630 190 660 070 570 070 60 10
2.36
2.3 Scenario Generator
2.3.1 Scenario Generator Module
As discussed in Section 2.1, subnetworks are functionally defined according to the
following applications: Command and Control, Field Artillery, and Air Defense. The three
subnetworks thus formed will be considered independent of each other, i.e., each subnet is
treated as an isolated entire network by itself. Some military units, e.g., Division Main CP,
may appear in multiple scenarios, and hence it is assumed such units have multiple packet
radio units. During the proposed 30-hour scenario, three deployments are considered; hence,
location of these PR units can be:
- Initial deployment
- Along DL ALPHA deployment
- Along DL BRAVO deployment.
Traffic can either be routed directly terminal-to-terminal or via a host where
messages are temporarily stored. The total traffic volume can also be modified by scaling
the volume per the 30-hour period; this is done for all needlines, but cannot be done for
individual needlines. Also, the length of time to be simulated must be specified. This time
depends primarily on how much time is required to achieve steady-state PR network
operation. In summary, specification of the subnet, deployment, host, scale factor, and time
duration defines a "scenario."
For any one scenario, the Scenario Generator module will generate the following input
files for PRSIM:
- An exogenous traffic profile
- Mobility profile
- PR terminal file.
2.37
i1aci_
The PRSIM module will then simulate the packet radio network operations with the
given PR unit locations (including mobility) and traffic as generated by the Scenario
Generator module.
2.3.2 Terminal Profile
The terminal PR location file output from the Scenario Generator includes theidentifying number of the packet radio unit, the coordinates of the packet radio unit for the
deployment, and antenna height. This information is directly extracted from the Data Base
module.
2.3.3 Traffic Profile
The traffic profile generated by the scenario generator and read by PRSIM specifies
the simulated exogenous traffic by each individual message transmission time. Exogeneous
traffic refers to traffic entering the network via terminals, which in turn generates internal
network traffic to transport external traffic. There is no internetwork traffic explicitly
considered in any scenario.
The simulated traffic file contents is sequentially arranged to their time sequence of
occurrence for the specified scenario. Figure 2.9 gives a sample scenario generator output.
The entries in the profile include the clock times at which the transmissions are scheduled,
origination and destination PR identifying numbers, intermediate PRU's required for
connectivity, and message lengths (in packets). Messages to be transmitted could be so long
as to exceed one information packet length. When messages comprise several packets, these
entries in TIME refer to the time the first packet is ready to be transmitted. Transmissions
of subsequent packets of the same message follow automatically in sequence in a chain of
events that follow transmission of a packet according to the packet radio protocol. The
ninth entry in the exemplary traffic profile shows that at time 4908910 seconds, device 1
(Main Command Post) has a 3-packet message to transmit to device 2 (Brigade 1). The
device number and description are associated from Data Base ( Tables 2.1, 2.2, and 2.3).
The source and destination PRU's are determined by the needline; the intermediate PRU'sare determined by shortest path routing. A standard shortest path algorithm was
* Implemented as part of the Scenario Generator to obtain these routes.
2.38
nlac.TIME ROUTE PACKETS
10 1 9 0 0 0 n 0 0 390410 9 1 0 0 0 0 0 0 3
264610-15 1 ..0-0-0---0.--0--0- 3179341n .1 3 4 0 0 0 0 0 3.2344010 1"" 1 5 0 0 v 0 0 3
- .-2409210- 8 15 1--0 -0--0..- 0- 0.- -3--2964510 4 A 1 0 0 n 0 0 33960lO 8 15 1 0 0 n 0 0 3
. 90 ql0. 11 10-2 -,0-Cl n- 0- .-.3.--4923210 1 9 0 0 0 0 0 0 35201510 1 11 10 0 0 0 o o 3
.-- 4-531610- 1 it. 0.-0.-0--0-.. 0-0-3--7075810 8 15 1 0 0 0 0 0 37574410 15 1 n 0 0 0 0 0 3
---. 7667610.-1 3 .O-.-0.-....- 0-- 0---3--78468410 1 14 0 0 0 0 0 0 3R008610 1 14 0 0 0 0 0 0 3
.537110- 8 15 1---..O--.-- 0--3-9696010 1 3 4 0 0 0 0 0 3
10048q1O 11 1 0 0 0 0 0 0 3.-- 10206610--1 -*l 5--D -0--a--0--.0--- -3---
11233410 13 1 0 0 0 0 0 0 31144 610 1 15 8 0 0 o 0 0- 312104210-13 --1 0-0 -. 0-0--0- --3...13184010 8 15 1 0 0 0 0 0 313191710 1 3 4 0 0 0 0 0 3
- 13300210-8 15 .---- 0---0-- ---0-- 0---3-13353610 11 1 0 0 0 0 0 0 3.23759410 1 9 0 0 0 0 0 0 3
--- 14058710--7 11 .1-0 -0-- O--0----3 --14909810 4 3 1 0 0 0 0 0 315252810 1 3 5 0 0 p 0 0 3
-- 15665510--9 -1 .0-0-0--f---n - 0---315q79510 1 9 0 0 0 a 0 0 3
FIGURE 2.9: SAMPLE SCENARIO GENERATOR OUTPUT
S 3.29
lac_For the C2 and Field Artillery terminals, the time between transmissions is assumed to
be random with an exponential distribution. If T is the clock time for the beginning of the
simulation time, then subsequent transmissions from that PR are:
T+ &l,T+ F,+ C2 ,T+ +E
where &I, &2' C3 are exponential -random variables generated around the mean inter-
transmission time rate for that PR. T is an arbitrary time for the beginning of simulation
and is identical for all devices in a particular scenario.
For ADA, the transmission times are not random but periodic. In this ease, the
interarrival times are assumed constant not random. Suppose D is the fixed time between
messages and T is the time of the first transmission, then subsequent transmissions occur at:
T + D, T + 2D, T + 3D, T + 4D ....
To compute the mean transmission time for a particular PRU, let
A i =number of transmissions from that particular PRU to PRU i during the busy
period.
The quantity X i is computed by dividing the 30-hour transmission rate by 1.25 to determine
the daily transmissions and then multiplying by the busy hour fraction to determine the peak
hour value. Let D be the set of all destination PRU's to which a single PRU transmits.
Then, the total transmission rate for that PRU to its associated destination group D is:
iCD
the probability of the transmission being destined for destination j in that group,
f = jfJ - -7
Then the mean lntertransmission time is 1/ x. For C2 and Field Artillery, the Scenario
Generator would then generate times for transmission
2.40
iicti T + Zr
k=1 k
as described above. For each t,, a destination PRU would be determined randomly using the
probability distribution fj, j D.
When a message-switching host is employed, the Scenario Generator automatically
redefines the needlines to accommodate the host. This is done as follows. Suppose PRU A
is a host and consider a needline from PRU's B to C, where A + B, A # C. Then the
Scenario Generator defines needlines:
- from Bto A
- from A to C
with the same parameters as in the original B to C needline. Needlines for which either
endpoint is a host are not modified. With the new needlines, the same methodology as
described above is applied.
For ADA, the same methodology would be employed, but the time between trans-
missions would be fixed, i.e., the probability distribution is discrete with a single point
having nonzero probability.
The following message length parameters were employed:
- 304 bits in the Field Artillery
- 5,640 bits in C2
- 76 bits ADA.
In this study, the message sizes are being taken as fixed for all information messages.
However, if a message length distribution were specified, varying message lengths could
have been simulated. The evaluation of the expected message length distribution indicates
there is a small variance in the distribution, so the fixed length assumption was deemed
valid.
2.41
A standard maximum packet length of 1,856 bits is assumed. Thus, the message of
5,640 bits would require four packets of 1,410 bits each. However, the 5640 bit message
size assumes a 12-bit Hamming character code is used. With the reliable communications
channel provided by the PR network, such a code is unnecessary. Assuming 8-bit characters,
there are only 3,760 (8/12 x 5640) bits of information. Thus, the information bits could be
accommodated in two 1,880-bit packets or three 1,254-bit packets. Three packets were
assumed.
2.3.4 Mobility Traffic Profile
The motion of the Aerial Observers will be simulated by PRSIM in discrete increments.
The initial locations of the Aerial Observers will be specified in the terminal location file.
Then, subsequent movements will be given in the Mobility Traffic file, as shown in
Figure 2.10. For each mobile PR unit, it lists PRU identifying number, the clock times (in
column TIME) when it moves from one sector of the quadrangle track to the next, and the
coordinates of the new sector.
There are only three Aerial Observers and they are in the Field Artillery subnetwork
only during the initial deployment.
2.4 The Link Module
2.4.1 LOS Link Model
MITRE Corporation determined which devices had LOS connectivity; this information
was incorporated into the data base.
2.4.2 Longley Rice Model
The alternative Link Module used to determine the communication links was based on
the Longley Rice program provided by CORADCOM. The Longley Rice [ 6 1 model computes
the attentuation of transmission between transmitting and receiving nodes. The model
employed was the enhanced Longley Rice model that accounted for location variability using
a stochastic algorithm. This algorithm is described in the Appendix. Then considering
relevant geographical and environmental conditions, together with transmission R/F and
* 2.42
NEW COORDINATESPR UNIT TIME IN KMS
79 820 X1 Y1
81 1,891 X2 Y2
80 3,160 X3 Y3
FIGURE 2.10: MOBILITY FILE FROM SCENARIO GENERATOR
2.43
antenna heights parameters, a minimum threshold value of energy of reception is specified,
below which a receiving PRU cannot "hear" the transmission. The Longley Rice model
program computes the attenuation loss between two PR units. The calling module can then
determine whether the two PR units can form a link, i.e., the attenuation losses still leave
enough transmission intensity to allow reception.
CORADCOM provided the attenuation thresholds of 150 dB at the 100-Kbps data rate
and 144 dB at the 400-Kbps data rate.
2.5. Packet Radio Simulation Module
2.5.1 PRSIM Capabilities
Because of the complexity of PRSIM, it is beyond the scope of this report to provide a
detailed description of its capabilities. However, in this section, a high level description of
PRSIM's capabilities and its input parameters is presented.
PRSIM is an event-driven simulation program. The exogenous events driving the
simulation program are the message arrival events as defined in the traffic profile,
generated by Scenario Generator module. PRSIM simulates both PR link protocol and end-
to-end protocol (TCP). However, the opening and closing of virtual circuits are not
simulated; in effect virtual circuits are assumed to be permanent.
The level of detail in PRSIM is exemplified by the basic events in the simulation; these
events are
- Message arrival
- Begin reception
- End reception
- End processing of received message
- Begin transmission (radio and hardwire)
- End transmission (radio and hardwire).
2.44
nac._Both radio and hardware transmissions are simulated. Three types of packets are
modelled in the simluation:
- Information packets
- End-to-end acknowledgments
- Local repeaters-on-packets (LROP's) (optionally).
Information packets are generated by the exogenous message arrival (corresponding to
the message arrival specified by the traffic profile). End-to-end (ETE) acknowledgments are
generated internally in the network when an information packet is received at the
destination (dependent on the TCP rules for sending such an acknowledgment). When a link
acknowledgment cannot be effected by echoing, PRSIM generates an active link
acknowledgment for both information and ETE acknowledgments.
LROP's packets constitute monitor and control traffic that is internal to the network.These packets are randomly generated for each PRU according to a rate and packet length
as specified in the parameter input.
The capture mode simulated by PRSIM is the half amplitude mode. A packet will be
received without interference if
- The channel is idle when packet reception begins.
- No packets of higher energy are subsequently received during the packet
reception.
However, packets received without interference may have bit errors. The effect of
detection and correction of such errors is simulated knowing the probability of a packetbeing received correctly or the probability of being able to correct a packet. These
probabilities are an input parameter.
2.5.2 PRSM Input
The PRMM module will have two sets of inputs:
2.45
I Iac- Scenario Generator output information
- User information.
As discussed above, the Scenario Generator output consists of:
- The terminal location data base
- The traffic data base
- The mobility data base.
The important user supplied operational parameters for PRSIM consist of the
parameters listed below:
- Length of headers in information packets (radio and end-to-end)
- Number of transmissions of an active echo in the network
- Maximum information packet length
- Fraction of high data rates capture loss due to not acquiring synchronization
- Fraction of low data rates capture loss due to not acquiring synchronization
- Number of buffers in station (based on PRU configuration)
- Number of buffers in a repeater (based on PRU configuration)
- Reactivation times of packet not acknowledged (based on TIU protocol)
- Retransmission times (based on PRU protocol)
- Number of retransmissions (based on PRU protocol)
* 2.46
- Number of reactivations (bsed on TIU protocol).
In addition, all simulations will be performed with a fixed PRU transmission power,
and no internetworking (host) traffic will be explicitly modeled. The numerical values used
for these parameters are specified in Section 3.
: 2.47
3. SIMULATION SCENARIOS
In this section, the inputs and the outputs of the simulation system are described. The
inputs consist of the variables that can be controlled while the outputs consist of the
statistics repeated by the simulation program.
3.1 Simulation Variables
3.1.1 Deployment/Subnetwork
The Command and Control subnetwork as deployed at the beginning of hostilities was
studied in detail. The Main Command Post was recognized as an obvious bottleneck; hence
initially smaller subnets around the Main Command Post was studied. The following two C2
subnetworks were
- Support scenario depicted in Figures 3.1 and 3.2
- Brigade scenario subnetwork depicted in Figure 3.3.
The Support Scenario consists of the following eleven PRU'S
- Main CP
- Division Support Command
- Maintenance BN
- Finance Company
- Supply and Transportation BN
- Adjutant General Company
3.1
LLI)
f%.3
I2:
0OC
CD,
9-4
C3,
CY-
1ijai6c W
I- U
Thr
co Z0..
CL
9-lC-N
*l 3.
83 DS BN1-40r 400 2 -t.40330 610
116 ADA BN1-441 84 DS BN1-41
20 DISCOMGSB14
8 MN BN1-BOIN
330 43D 0 NBI8I
1000
11 MN BN1-2AR
FIGURE 3.3: BRIGADE SCENARIO (LOS LINKS)
S 3.4
Aviation BN
Communications, Electronic Warfare and Intelligence BN
Medical BN
Military Police Co.
- Engineering BN.
These PRU's were initially chosen for simulation because of their proximity to the Main
Command Post. Then the Brigades were chosen for simulation with the associated support
battalions. The 15 PRU's in this simulation consisted of
- Main CP
Division Support Command
Field Artillery Group
- Division Artillery
- Air Defense Artillery
- Ist Brigade
- 2nd Brigade
- 3rd Brigade
- Direct Support BN 1-40
- Direct Support BN 1-41
3.5
Direct Support BN 1-42
General Support BN 1-43
- Maneuver BN 1-80
- Maneuver BN 1-82
- Maneuver BN 1-2AR.
These PRU's were chosen because of thier obvious functional importance.
Then nearly the entire Command and Control subnetwork was studied (only 4 PRU's
were excluded).
3.1.2 Connectivity
As discussed in the Appendix, the calculation of attenuation loss is a difficult problem.
For example, multipath, foliage, and terrain effects must be considered. Hence, it is very
difficult to predict the network topology a priori. Two approaches have been used
- Random Longley Rice
- Line of Sight.
The random Longley Rice, which is described in the Appendix is an empirical model
that computes attenuation given a measure of the terrain irregularity, h, the interdecile
range.
The major drawback of this model is that it is based on averages and thus cannot take
into account the detailed terrain profile. Hence, an LOS model was also used; in this model
it was asumed that two PRU's could communicate with each other if and only if the two
devices were in LOS of each other. Mitre Corporation performed the detailed terrain
profile analysis to determine which devices were in LOS of each other. In all cases, the free
qace attenuation was less than the 150 dB threshold; hence, it was assumed that a link
existed between all devices which were in LOS of each other.
3.6
Ilac._3.1.3 Message-Switching Host Requirements
The Command and Control application could be implemented in either of two ways:
- Store and forward through a message-switching host
- Terminal to terminal.
In the store-and-forward mode through a host, the origination terminal sends its message to
a host. Here the message is stored until the destination terminal requests the delivery of
the message. The request for delivery is not included in the model; also messages from the
host to the destination terminal informing the destination terminal a message is ready for
delivery are not included in the model. Thus in this model, any message between terminals
(not including the host) results in two messages, one to the host and one from the host. Thus
the traffic volume is essentially doubled by the introduction of the host. Furthermore, the
traffic is centralized at the host, thus additional congestion is created. The Main Command
Post was selected as the host. The possibility of employing other PRU's as hosts was
considered (both instead of and in addition to the Main CP), but there were no other logical
choices. Since adequate performance was obtained with the Main CP as the host, this issue
was not further addressed.
In the terminal-to-terminal mode traffic is transmitted directly from the origination
PRU to the destination PRU. Packets of the message will be stored in the intermediate
PRU's as necessary, but the entire message is not stored in an intermediate host.
3.1.4 Route Selection
As discussed in Section 2, routes are assigned a priori in the Scenario Generator using
a shortest path criteria. If a message-switching host is used, shortest paths are determined
between the origination/destination terminals and the host. For the terminal-to-terminal
mode, shortest paths between the origination and destination terminals were determined.
For the simulations using the random conneetivity model, special procedures to
minimize cogestion were designed; these procedures are described in Section 6. Because a
smaller number of hops were required for LOS and terminal-to-terminal models, it was not
necessary to develop such procedures in these cases.
3.7
3.1.5 Traffic Sensitivities
The traffic needlines, volumes, and message lengths are those supplied by MITRE as
described in Section 2. The PR network offered rate is defined as
P L
where
X = message arrival rate
L = message length (bits)
R = data rate (bps).
Theoretically, because of spatial diversity in the terminal-to-terminal case, the network can
support an offered rate of more than one. For example, two pairs of terminals that are
geographically separated can simulaneously communicate, and thus, throughput is not
bounded by one. However, in practical cases, the offered rate is typically a small
percentage.
Initially, the message arrival rate was chosen as the nominal rate derived from the C2
data base. Subsequently, this was scaled in the study by a factor of 2. The message length
and data rate were initially held fixed.
3.1.6 PRU Parameters
The parameters in the PRU were primarily based on the experimental packet radio as
defined in the Channel Access Program (CAP) 4.6. These parameters are discussed below.
3.1.6.1 Buffers
The number of buffers in the PRU's was initally chosen as 6, the number of buffers in a
PRU repeater. Subsequently, the number of buffers was reduced to 3. The simulation did
3.8S
ilac_not have the capability of functionally allocating buffers, i.e., allocation of certain buffers
for transmision or reception or radio or 1822 use.
3.1.6.2 Link Level Protocol Retransmissions
When the packet radio transmits a packet on the radio channel, it awaits a link (hop-
by-hop) acknowledgment. The PRU times out for a random period of time and will
retransmit the packet if the link acknowledgment is not received. The number of times the
packet will be retransmitted is a parameter.
The following parameters were used in all simulation runs:
- Expected timeout awaiting link acknowledgment: 9 milliseconds
- Number of retransmissions: 4
- Header size (including CRC and preamble): 256 bits.
3.1.7 End-to-End Level Protocol
Analogous to the link acknowledgment in the PRU, the TIU will retransmit packets if
end-to-end acknowledgments are not received. For clarity of presentation, such
retransmissions are referred to as reactivations; thus retransmissions occur at the link
protocol level and reactivations occur at the end-to-end protocol level.
The specification of the reactiviation parameters are based on the transmission
Control Program (TCP) Version 4. These parameters are:
- Number of reactivations: 10
- Random timeout awaiting receipt of end-to-end acknowledgment: 0.6 seconds
- Window size for a virtual connection: 5 packets (785 bytes)
- Header size: 150 bits.
3.9
1iac.Message lost in the PRSIM refer to those messages that were discarded after the maximum
number (10) of reactivations were exceeded. Because of the communication between
terminals consists of small messages interspersed relatively far apart, the window size did
not affect packet transmission. The opening and closing of virtual circuits was not
simulated.
3.1.8 Access Method
The non-persistent Carrier Sense Multiple Access method was used in all simulations.
Because adequate performance was obtained with CSMA, other alternatives were not
explored.
3.1.9 Capture
Half-amplitude capture was used in all simulations. A packet is received correctly in
this capture mode if and only if:
The receiving PRU was idle when the transmission begun.
No packets of higher energy were subsequently received during the packet
reception.
- There were no bit errors in the packet.
3.1.10 Packet Errors
Packets can be received in error if the
- Receiving PRU does not acquire or maintain synchronization upon packet
reception.
- Cyclic redundancy checksum (CRC) does not match.
3.10
The probabilities of not acquiring sync or not maintaining it are typically small and
were ignored in this study.
The CRC check is modeled via a channel bit error rate. The packet error rates
assumed were
- 1% (initially)
- 10%.
Since these error rates are based on a 1000-bit packet, a linear scaling is performed to
estimate packet error rates for other size packets. Thus smaller packets have a lower
probability of being in error than larger packets.
3.1.11 PR Processing Times
Upon receipt of a packet in the radio channel, the processing delays for error
correction (if applicable, BER greater than zero), protocol execution, and routing are
simulated. The processing values assumed are:
- Processing delay: 6 msec
- Error correction delay: 4 msec.
3.1.12 Monitor and Control Traffic
The monitoring and control traffic in the PR network is simulated by the transmission
of Local Repeater or Packets (LROP's). These packets are generated randomly with uniform
distribution having the mean inter-transmission time of 60 seconds for each PRU.
Other monitor and control traffic such as Performance Data Packets, diagnostics
(XRAY), route requests (to a station), or route assignment (from a station) were not
modelled.
3.11
Si
iiac_3.2 Statistics
In this section, the primary measures of system performance measured in PRSIM are
described.
3.2.1 Delays
PRSIM has the capability of measuring message and packet delays on both a one-way
and round-trip. The mean and standard deviation of following quantities are measured:
- First packet delay - time interval from the instant the origination TIU is ready
to transmit the packet until the destination TIU receives the last bit of the
packet.
- First packet round-trip packet delay - time interval from the instant the
origination TIU is ready to transmit the packet until the origination TIU receives
the last bit of the ETE acknowledgment for that packet.
- Round-trip message delay - time interval from the instant the origination TIU
is ready to transmit the message until the origination terminal receives the last
bit of the ETE acknowledgment of the last packet of the message.
- One-way packet delay - time interval from the instant packet transmission is
begun by the origination TIU until the last bit of the packet is received at the
destination TIU.
- One-way message delay - time interval from the instant the origination TIU
transmits the first bit in the packet until the destination TIU receives last bit of
the last packet of the message.
The first three quantities are recorded by the number of hops the packet must traverse
from source to destination. Hence, delays per hop were computed for these statistics.
3.12
nac_3.2.2 Offered Rate and Throughput
The offered rate, Of, is a normalized measure of the traffic volume generated
exogenous to the network; it is defined
nmLOf -
where
nm = Number of messages arriving during the simulation time duration.
Lm = Message length in bits (information bits only).
DR = Radio channel data rate.
t = Time duration (in seconds) of simulation
Typically, Of is expressed as a percentage and is less than 100%.
The throughput is a normalized measure of the packets that have been reliably
delivered by the PR network; it is defined as
npLT pm
where
n = Number of packets that have been delivered to the destination and for
which end-to-end acknowledgments have been received.
Lm = Packet length (information bits only).
DR = Radio channel data rate.
Typically, both the offered rate and throughput are expressed as percentages and are
much less than 100%. This bound on throughput is strict when packets are routed through a
3.13Sl
host. However, for direct terminal-to-terminal routing, the throughput can exceed 100% by
taking advantage of spatial diversity.
3.2.3 Buffer Utilization
Buffer utilization is the fraction of time the buffers in the PRU are occupied. The
occupancy time is measured from the instant that the packet is correctly received until it ispurged (due to receipt of hop-by-hop acknowledgment or maximum number of transmissions
is exceeded).
Buffer utilization was not measured, but buffer overflow and maximum use were
measured. Specifically, the number of times a packet had to be discarded because no buffer
was available is measured and the maximum number of buffers used.
3.2.4 PRU Radio Packet Transmission/Reception Statistics
PRSIM measures the following statistics for the PRU radio interface
- Packets captured
- Packets not captured
- Packets received incorrectly (bit error)
- Packets received correctly
- Number of first transmissions
- Number of retransmissions.
These statistics provide an indication of the PRU activity.
3.14
4. SIMULATION RESULTS
This section contains the results of simulation of the:
Brigade scenario
Entire C2 network scenarios
Support scenario
as they were defined in Section 3.
In Section 4.1, the highlights of this deployment study are summarized. Results are
then presented in detail in the sections characterized by the scenario and traffic profiles
considered as follows:
Section 4.2.1 Brigade Scenario Point-to-Point Routes LOS Links
Section 4.2.2 Brigade Scenario Host Routes LOS Links
Section 4.3.1 Entire C2 Scenario Host Routes LOS Links
Section 4.3.2 Entire C2 Scenario Host Routes LOS Links
Section 4.4 Support Scenario Host/Point-to- LOS and
Point Routes Random Links
The detailed statistical results are presented for each of the above cases in Tables 4.2
through 4.6. In the subsequent text, the referenced Run Numbers refer to the run number in
Column 1 of these tables.
4.1
4.1 Summary
The simulation studies performed for the Command and Control deployment of PR
networks in the Fulda Gap area indicate that the C2 traffic load can be handled by PR nets
over a wide range of operating conditions. Mean round trip (end-to-end) message delays of
less than 2 seconds in the worst case can be measured. This includes the increase of traffic
of 100% over the estimated volume and the reduction of PRU buffers from a nominal value
of 6 to 3. These conclusions hold for both host and direct (terminal-to-terminal) routing for
nearly all cases. The only exception is the host case with twice the nominal traffic load and
three buffers, as discussed in Section 4.3.2. However, the performance of point-to-point
routing is superior. Hence, operation of point-to-point routing could support a large traffic
volume.
The introduction of dedicated repeaters in the forward area near the Bridgade PRU's
reduces the number of hops certain messages must traverse in reaching their destinations.
This results in a corresponding decrease in the message delays. Hence, even in cases where
there is sufficient connectivity to guarantee a path exists between any pair of PRU's,
network performance can be improved by the incorporation of dedicated repeaters. This is
expected because the offered traffic loads are small.
The retrograde from the initial deployment to the DL BRAVO deployment results in a
decrease in network connectivity. However, network performance (as measured in the
Brigade Scenario) does not deteriorate and delay requirements can still be statisfied.
Typically, messages had to traverse more hops, but maximum delays were less than 1.5
seconds.
In the sensitivity analysis, the size of the PRU buffer was found to be the most critical
parameter. The traffic volume was determined to be extremely light, even after the traffic
was scaled by a factor of 2. The low intensity of the offered traffic is illustrated in the
following analysis. It is assumed in the generation of the traffic that interarrival time
between messages is Xand t is the time required to transmit a message to its destination
and receive an end-to-end acknolwedgment. For point-to-point routing, X equals 797
messages/hour; while for gateway routing, X equals 1,173 messages/hour.
Then,
P = P (no arrivals in (h, t + h)) = e" t
lIl4.2
iiacmfor any message arrival time h. For the host requiring routing, Pt equals 0.72 for t = 1,
implying that with probability greater than 0.72 a message of less than 1 second is
completed before the next message arrives. The corresponding probability with gateway
messages is 0.80. Other values are tabulated in Table 4.1. The general implication of this is
that at any particular time there are relatively few messages in the network. This was
observed in the simulations.
4.2 Simulation Results of the Brigade Scenarios
An initial set of runs were made with the brigade scenario, comprising of a subset of
the C2 units which were relatively more important. With the obvious high traffic to and
from the DIV MAIN PRU node, this subset represents a high traffic subset of C2 . The
brigade scenario simulations were performed in two sets. The first set used traffic profiles
generated for point-to-point routing. The second set of simulation runs were with traffic
profiles as generated with DIV MAIN operating as a host node for each message.
4.2.1 Simulation Results of the Brigade Scenario with Point-to-Point Routed Traffic
Initially, the simulation (Run 1) was run with traffic generated at nominal traffic load
as recommended by MITRE. Twenty-five minutes of real-time network operation was
simulated; this resulted in the transmission of 117 messages. Round-trip packet (first
packet only) and round-trip message delay means varied from 0.156 see and 0.338 see for the
shortest paths (one hop) to the worst case of 0.176 seconds and 0.657 see for the longest
routes (four hops), respectively. From the results of these experiments, it was concluded
that the brigade scenario subnetwork could more than adequately handle the present traffic
load.
To test the subnet more severely, a two-step approach was then taken. As a first step
(Run 2), the traffic input rate to the network was scaled up 100%. As a second set (Run 3),
with this increased traffic, the number of buffers available to each PRU was halved from 6to 3. With the traffic doubled, there was relatively little deterioration in performance; the
delays for the longer routes increased comparatively more than for shorter routes. This
increased delay is typically reflected in the increased round-trip message delay mean of
0.413 see, up 9.8%.
The subsequent reduction in buffer capacity introduced caused more significant
changes in delay. One-way message delay was now 0.489 seconds (up 18%), and on a per hop
4.3
TABLE 4.1: PROBABILITIES THAT MESSAGE SERVICE COMPLETEBEFORE NEXT ARRIVAL
Probability Message is CompletedBefore Next Arrival
MessageService
Time (See) Host Point-to-Point
0.5 .85 .89
1.0 .72 .80
2.0 .52 .64
4.I 4.4
lea~
0 ~ 1c; I Iil C
94 #41 C 0
M USDCD t- -0 to to , t-W W - q4 o
c; 0
C 0000 0@6 aC o 0 00
- CD
00 A0
.CD 0.
A. A.C,. . c o
'It .- ;C D C
C; =i=C DCDC D
.; . CD C! C; j .; C0C 0 CCC CDC0 00 CDO O
INC , , ,
.9E 94 C 4nA.
a. -
~G1De
A. CC~~ C1#4~ *t~ P1.i~44@
basis, measured round-trip message delays were up an average of 15%. There was no
indication of messages with longer routes being especially delayed. The general increase in
delay times could be associated with the frequent unavailability of buffers, causing the
relatively higher ratios of retransmission.
In an attempt to control excessive copies of packets in the subsequent run (Run 4),
with traffic and buffer capacity held constant, a flow control procedure was implemented.
This was done by doublng the time between terminal transmissions. This improved delays
generally on a per hop basis. Also, a general reduction in buffer congestion was observed.
However, the introduced delay in sending out the information packets caused increased in
one-way and round-trip message delays of 2.5% and 27%, respectively.
The Brigade scenario was next simulated (Run 5) with the same increased point-to-
point traffic and six buffers but in the DL BRAVO deployment. The DL BRAVO deployment
has a new set of LOS links. With the LOS link connectivity, the needlines of the scenario
are now satisfied by routes varying from a minimum of one hop to a maximum of four hops.
The mean number of hops goes up to 1.87 in the DL BRAVO deployment as compared to 1.33
for a similar run in the initial deployment.
The overall round-trip packet and message delays, however, do not change
significantly. Thus, the round-trip packet delay per hop is 0.158 seconds which is 1.3%
higher and round-trip message delay per hop is 0.300 seconds which is only 5.6% less than
their corresponding values in the initial deployment. Even with its greater mean number of
hops in the DL BRAVO deployment, the overall delay means are comparable to the initial
deployment values and can be further illustrated by looking at one-way delays. It is observed
that in the worst case (i.e., longest routes) for initial deployment one-way packet delay is
0.182 see. The one-way delay for the longest routes in the DL BRAVO deployment is only
0.170 sec, an improvement of 7%. In fact, overall one-way message delay is also 0.351 sec
which is an improvement of 15%. This change in delay is likely caused by the change in
network topology from the initial deployment to DL BRAVO.
From the above, we conclude:
- Traffic loads are significantly lower than the subnet capacity. Thus, significant
changes are not noticeable even by doubling traffic.
- At the operating traffic loads, while six buffers per PRU may be slightly more
than optimally required, at traffic loads of this order, it is a sensitive factor.
4.6
A buffer reduction will cause greater message delays. However, the congestion
introduced does not catastrophically deteriorate network performance.
Even with a significant decrease in network connectivity in DL BRAVO, network
performance is still adequate.
4.2.2 Simulation Results of Brigade Scenario with Host Routed Traffic
As explained in Section 3, other traffic generating parameters being the same, a
traffic profile with routing through a host node is approximately twice the number of
messages for a similar but point-to-point routed traffic. With DIV MAIN operating as
centrally located host, the shortest paths for needlines are shorter than for point-to-point
routes. However, owing to the distribution of number of hops per needline, the overall mean
number of hops/message remains essentially the same around 1.36. Furthermore, the effect
of buffer capacities would be expected to be more critical at the host and possibly other
close by heavily loaded nodes.
As in the point-to-point case, for purposes of comparison, a base run (Run 6) was made
with the nominal traffic and six buffers per PRU. This comprised of 166 messages over the
25 minutes of real-time simulated. Subsequently, in a two-step process of introducing
congestion, the buffer capacity was halved and then traffic was increased by 100%.
With initial traffic and six buffers (Run 6), round-trip packet and message delay means
varied from 0.165 see and 0.341 see for the shortest routes to 0.296 see and 0.522 see,
respectively, for the longest routes. These delays are comparable to their values in the
comparable point-to-point run. Although there is increased congestion at the host, overall
delays are comparable because the load is relatively light.
Buffer reduction (Run 7) caused a start in buffer overflows and higher incidence of
retransmissions. Generally, these caused higher delays (e.g., round-trip message delay mean
up 6% to 0.317 see), but with traffic still being relatively light the effect was not
significant. In the second step (Run 8), introduction of twice the traffic, significant changes
were seen.
- Round-trip message delay up 32% to 0.418 see
- One-way mesage delay up 65% to 0.539 see
0 4.7
aa
40 ~ ~ ~ ~ 04 ut - iWt a w 1 4 !t-W t- "aa = o
S 0-D
SO .C!5it
a,0g
(a 0 0 L- CO WPb11 :"- 0 a a DCa -: C, 41C; C;W C;C; a C;
9.4 C4 0
I-V0 ~~ I - - ol-
44~, --aqPb4
Round-trip packet delay up 21% to 0.185 see/hop.
In observation similar to the point-to-point case, it was observed that a significant
number of buffer overflows had increased with an appreciably higher ratio of transmissions.
However, delays were still of a reasonable magnitude, and the network was capable of
taking the load.
In order to increase buffer availability with the same traffic and network conditions,
the packet error processing time was eliminated (Run 10). Thus, packets received in error
would not occupy buffer space awaiting further copies to be used in their error correction.
This resulted in a 5% improved round-trip message delay time. This was effective in
increasing buffer availability because many packets required in a PRN are extraneous; also,
with a message error rate of .01, multiple copies of a packet are typically not needed for
error correction.
The effect of introducing monitor and control traffic (LROP) was also investigated.
Initially, a simulation (Run 9) was performed with the nominal traffic value and buffer
capacity, but with LROP packets being transmitted. The resulting delay statistics were not
markedly different from those measured in the ase run. In fact, all variations were within
4%.
Then, a simulation (Run 11) was made with the scaled up traffic, three buffers and
LROP's. In the run, the probability of receiving a 1,000 bit packet correctly was reduced
from the value of 0.99 to 0.90. As would be expected, compared to the comparable run with
probability 0.99, delays now were greater and had significantly higher standard deviations.
This wide disparity is due to the now greater number of packets which are received
incorrectly, and thereby, have to wait for retransmissions (copies) to get through. In fact, a
large number of LROP's can be received incorrectly and tie up buffers. With the increase in
the number of packets received incorrectly and limited buffer capacity, the number of
overflows were about three times more, and PRU's had significantly higher ratios of
retransmissions to first transmissions. In numerical terms, the highlights of changes
(computed relative to simply reducing buffer capacity and increasing traffic) were:
- Round-trip message delay mean/hop up 100% to 0.829 see
Round-trip packet delay mean/hop up 48% to 0.273 see
4.
Total round-trip delay mean was 1.31 see in the longest routes (2 hops) and total
round-trip packet delay times for these routes werw 0.506 sec.
Since the introduction of LROP's in previous experiments had only marginal effect, we
would conclude the increase in packet error rate is much more significant than the
introduction of LROP traffic.
Finally, the brigade scenario was simulated (Run 12) with gateway traffic up 100% and
employing 6 buffers per PRU in the DL BRAVO deployment. In the DL BRAVO deployment,
the new LOS links provide less connectivity than in the initial deployment. Thus, to route
host traffic, the routes vary now from one to four hops long as against only one or two hop
routes in the initial deployment. This translates to an increased mean number of hops to
1.66 in this deployment as compared to 1.35 for the initial deployment.
Generally comparing results of the run, it is seen that for the longest routes (4 hop),
one-way delays are comparable and sometimes even better than the corresponding delays for
the longest routes (2 hop) in the initial deployment. However, congestion at and around the
gateway seems to cause increased round-trip message delays, the worst case being 0.780
see.
It is seen that delays for 3-hop routes are greater than the corresponding delays for 4-
hop routes. This further implies that in this scenario route lengths do not completely
influence delays. Rather, connectivity at and around the host has a more important impact.
From the above results, we conclude:
- At nominal traffic loads, host traffic does not introduce sufficient congestion to
cause appreciably increased delays over the point-to-point routed case.
- With host traffic, a reduction of buffer capacity has a more pronounced effect
on delays than in point-to-point routed traffic. It is possible to isolate certain
PRU nodes in the entire C2 net that may show their relative locations, be
particularly susceptible to reduced buffers.
Elimination of packet error processing time in PRU buffers enables smoother
and faster traffic flow.
The effect of introducing LROPIs as additional traffic at the initial level of
traffic does not affect network characteristics.
4.1
The environment (quantified by the assumed message error rate) has a very
significant effect on delays.
It appears due to a high connectivity of the PRUIS in this scenario, routes are
reasonably short. This allows grood traffic flow and helps keep the subnet from
degenerating under more severe conditions of buffer availability, increased
traffic and lower probabilities of receiving packets correctly.
Even with a significant decrease in connectivity in DL BRAVO, network
performance does not deteriorate.
4.3 Simulation of 32 PRU's Net Scenario
In this study, the performance of a PR net with 32 PRUIS was evaluated. In this case,
only four PRU's (each with a low traffic generation rate) were excluded from the scenario.
The excluded PRU's were:
- Cavalry Squadron
- Engineering Battalion
- Maintenance Battalion
- Finance Company.
4.3.1 Simulation Results of C2 Scenario with Point-to-Point Traffic (32 PRUs)
As in previous sets of runs, a base run (Run 13) with the nominal peak rate of traffic
and six buffers were made comprising of 289 messages during the 25 minutes of real-time2
simulated. Because of a greater spread in location among the C PRU's compared to the2
Brigade scenario, point-to-point routes for C scenarios needlines vary from one to four
hops. For example, the communication of Maneuver Battalions with the Main Command
Post typically requires a large number of hops. The mean number of hops is roughly 1.72 as
compared to 1.35 for the Brigade scenario. Round-trip packet and message delay means In
* 4.11
C! 'I. I '
_p a
4) CD 00-0 00 4 0 00 0
o-l a
W! Sn nn Li A
eq o eq 0 C
1101- R10 C!o -:1 501 10
C IVoo. ~ ~ ~ ~ t
aI-R ~t~q
4.12
na,_this scenario were 0.159 see and 0.347 sec for one-hop routes to 0.238 see and 0.985 see,
respectively for the longest routes (4 hops). Hence, all one-way and round-trip delays
(actual) were appreciably greater than for the similar runs with brigade scenario.
Subsequently, runs were made by:
- Increasing traffic 100% but keeping buffer capacity at 6
- Decreasing buffer capacity to 3 but keeping initial traffic constant
- Both increasing traffic 100% and decreasing buffer capacity to 3.
The effect of increased traffic (Run 14) was only marginally greater in heavier traffic
case. For example, the one-way message delay increased from 0.526 sec to only 0.589 when
the traffic volume was doubled. The round-trip message delay increased from 0.985 sec to
1.103 sec. Furthermore, on per hop basis round-trip packet and message delays/hop were
0.163 sec (up 1.2%) and 0.322 sec (up 1.3%), respectively.
The effects of a buffer reduction (with traffic scale factor 1.0) were more pronounced
(Run 15) than those of increased traffic. Thus, round-trip packet and message delaymeans/hop were 0.172 sec (up 7%) and 0.403 sec (up 27%), respectively. It was observed
that delays for the relatively longer routes were especially higher. This is due to buffer
congestion. Thus, the longer a packet route is, the more times it has to compete for buffer
space in PRU's in route. The run with buffer size 3 and nominal traffic yielded round-trip
packet and message delays of 0.582 see and 1.376 sec for the longest routes. These figures
are increased of 6% and 40% over their values in the base test run.
In the next run (Run 16), the effects of reduced buffer capacity and increased traffic
were combined. Additionally, LROP packets were introduced with regular traffic. The
combined effect produced delays of almost 2 sec which appears close to the assumed
maximum acceptable value of round-trip message delay. In the best case, i.e., for the 1 hop
route round-trip packet and message delays, were 0.184 sec and 0.563 sec, respectively. In
the worst case of 4-hop routes, these values respectively were 0.345 sec and 1.863 sec.
A rather large number of buffer overflows occurred, and the general level ofretransmissions needed was high. It was observed that standard deviations for longer route
statistics were proportionately high. This would Indicate due to the lack of buffer space a
relatively large number of these meassages were reactivated in the TIU after not receiving
4.13
na._an ETE ACK.
In two separate runs, LROP traffic was considered with first initial level traffic (Run
17) and their doubled traffic (Run 18). In both cases, the effects of LROP's were not of
significance. Statistics reported in each of these runs were essentially the same as the
respective runs without the LROP's.
From the above results, we conclude:
- Traffic for C2 scenario is light and an escalation of traffic by 100% does not
produce significant variations in performance.
- Reduction of buffer capacity significantly deteriorates performance. For
example, buffer reduction causes relatively significant delays for 3 hop and
longer routes by as much as 40%.
- Under the dual effect of reduced buffers and increased traffic, C2 net scenario
produces delays that are around 2 sec for messages traveling 4-hop routes.
Traffic level is not as serious a problem as unavailability of buffers.
- With six buffers, LROP traffic did not cause any deterioration in performance.
4.3.2 Simulation Results of C2 Net Scenario with Host Traffic
In this scenario, DIV MAIN again operates as a host for all traffic. The shortest path
routes vary from 1 to 4 hop routes. The mean number of hops/message of 1.59, however, is
less than 1.74 as in point-to-point routing. As explained in Section 3, host traffic for a
scenario is approximately twice the number of messages for the same conditions with point-
to-point routing.
As in the previous studies, a base ru. (Run 19) was made with the recommended peak
rate of traffic and six buffers. Round-trip packet and message delays for this case were
0.162 see and 0.355 see for 1-hop routes compared to 0.869 see and 1.266 see (respectively)
for the 4-hop routes. For the 1-hop routes, these delays are essentially equal to those
determined in the point-to-point case, but for the 4-hop routes, there are significant
neres. For example, in the host case, the 4-hop round-trip packets (.869 see) and round-
tip aomqe (1.266 see) are 59% and 29% greater than the delays experienced in the point-
4.14
I !1; CD RCDI I I IIi
40 a".
I~~0I 1111
2 'F-
10.
F; C C ; ;C , 4 DC
#4
0z Ad
t- ,C.50 M0h4 -t
-C U 2t-1a.4 ~ ah t~
CD !z CD c; C;-, -. ,
co 0
U * U)A. eahU~a~ m~~ o#4C W
z~o~t 0 g~U O o~4o O~t4*g
caC CC CC~@ @@ i
8D 00 -, ob . , 0bCh4 DMi- wir ww c- b Cs 40 o
GoC :e l R1 :e De - lN 4c
I.3 -0 C4 C4V 4VS o a
C49
-o La0.1 X
2 ,0 0.m*3C VD
.3E. 3 -!#
CtC
to-point case. This is caused by the congestion around the host where PRUs (83 and 24)
experience a large number of retransmissions. Note the standard deviation in the host case
for messages with 4 hops with 0.850 see is high; this is likely caused by messages not being
successfully transmitted on the first attempt and hence requiring reactivation in the TIU.
The next simulation (Run 20) made was with the LROP packets as additional traffic to
the nominal host traffic employed. As had been experienced in the past with LROP's, at this
level of traffic, they introduce no effect on delay statistics. For example, delays per hop
were less than 1% different. The existing congestion around the host is once again
highlighted by the increased number of retransmissions necessitated by the LROP's.
In a subsequent simulation (Run 21), buffer capacity was reduced to 3 and the initial
traffic was used with no LROPs. As expected, buffer overflows frequently occurred with
the host having the most of them. There were more retransmissions as evidenced by the
retransmission ratios of device 12, increasing from 1.40 to 1.58. Also, the buffer overflows
caused appreciably higher delays. This round-trip packet and message delay per hop was
0.200 see (up 17%) and 0.436 see (up 30%), respectively.
Next, the entire C2 network was simulated (Run 22) with the host traffic scaled up by
a factor of 2 and using 6 buffers. The results were essentially as expected, compared to a
similar run, but with initial level of traffic, the changes were consistent with observations
made from other runs. Round-trip packet and message delays per hop did increase to 0.179
see (up 5%) and 0.345 see (up 3%), respectively. As mentioned with the point-to-point set of
runs for C2 net, it is similarly observed here also that the increase in delays with greater
traffic is modest indicating a light traffic load.
Finally, the scenario was simulated (Run 23) with the introduction of two relays in
locations and LOS links as follows:
Relay 1: 490 394
Relay 2: 500 210
The introduction of relays brought the mean number of hops per message down to 1.50
from 1.60 in the equivalent run without relays. Round-trip packet and message delays per
hop also Improved to 0.174 see (3% decrease) and 0.340 see (2% decrease). These
improvements in overall delays considering all messages of different routes seems small.
But the introduction of relays with LOS links to the host is estimated by looking at
4.16
Ilac_
the improvement they bring to the longer route delays. Thus, for the longest route, the
round-trip packet and message delays per hop are 0.504 see (16% improvement) and 0.872
see (23% improvement), respectively.
Subsequent experiments in which buffer capacity was reduced to 3 indicated severe
congestion around the host. Throughput deteriorated to zero indicating that the network
could not operate with only 3 buffers at the gateway.
From the above results, we conclude:
- The base run statistics compared to the similar run employing point-to-point
routing indeates that round-trip delays for 3- or 4-hop routes may be reduced by
employing repeaters to improve performance.
- At the traffic rates considered for PRU's comprising the gateway and ones with
LOS links directly to it, buffer capacity of 6 seems optimal. Any further delay
would result in comparably higher delays especially for longer routes messages.
- Simultaneous increase in traffic by a scale factor of 2 and reduction of buffer
capacity to 3 causes a catastrophic deterioration in performance.
4.4 Simulation Results of Support Scenario with either Host/Point-to-Point Routing
Also, a subset of PRU geographically positioned around the Main Command Post was
evaluated; this subset is referred to as the Support scenario. Three runs were made, all with
the same volume of traffic as recommended (nominal traffic loads). The variation of the
runs was in the routing employed and link criteria. LOS links were used for one run
employing a host (Run 25) and the other point-to-point (Run 26) routing traffic. In a third
run (Run 27), a randomly generated (using the Longley Rice model) set of links was used to
route traffic through host DIV MAIN.
In this scenario, the performance of the three cases is very similar. This is illustrated
by the following mean round-trip delays:
Host/LOS .328 see/hop
4.17
j ; ~
CD M -f
n. W! n0 00
9131
00 a
CCca
*R '!. U, CD 4C
66
E!J0 ; -1C 49 4s
z.1
8 * ~ 4c
EeHol
E 1W.
- Point-to-point .332 see/hop
- Host/Random .316 see/hop
The delays above are given on a per hop basis, i.e., the measured round-trip delay is divided
by the number of hops and averaged over all samples. Note that in the LOS cases, the
routes have a maximum number of two hops while in the random cases, there are some
routes with three and four hops. Thus, the longest round-trip delays are experienced in the
random case with routes having four hops. However, even in this case, the round-trip
message delay is less than .700 seconds.
On a comparison of the two host traffic runs for this scenario, it is observed that the
connectivity using LOS links yields a general improvement as compared to random set of
links determining connectivity. Thus, the round-trip packet and message delays per hop
reduce 5% (to 0.149 see) and 4% (to 0.316 see), respectively, but these changes are
relatively small and insignificant. However, it is pointed out in the preceding paragraph, the
random link connectivity results in an increased mean number of hops per route. This would
normally be expected to produce an increase in overall mean delays which is not reflected in
the results. It is inferred that there is a general congestion around the host. This is not
catastrophic at these loads of traffic. Nevertheless, the use of more varied routes and
different PRU's to reach the host reduces the delay.
Comparing the two runs using LOS connectivity essentially is a comparison of network
performance with:
roughly a double rate of traffic (actually closer to 50% since messages to/from
the host require only one transmission in both cases)
- centralized concentration of traffic at the host node.
The delay statistics of the two runs generally differ by a few milliseconds only. It is
observed, as would be expected, heavier congestion around the host and higher ratios of
retransmission needed by PRUs serving as access to the host. However, the effects of this
as seen are minor at these loads.
From the above results, it is concluded:
• 4.19
Inc_The support scenario connectivity allows for satisfying its needline requirements
with of the same order as the brigade scenario.
- Performance with LOS and random links is comparable.
- A certain small improvement in delay times may be effected by distributed
traffic through different PRU's to reach the host node. This is especially so for
longer routes. However, the changes are minor.
4.2o
0N
5. THROUGHPUT ANALYSIS FOR TANDEM REPEATER NETWORK
5.1 Summary and Introduction
In this section the maximum throughput is computed for a tandem network of N
repeaters depicted in Figure 5.1. Both Carrier Sense Multiple Access and ALOHA Multiple
Access techniques were employed. In this network, repeater i can communicate with only
repeaters i - 1 and i + 1. Each repeater (2,...,N-1) wishes to send messages at a rate S in
each direction, while repeaters 1 and N transmit in one direction only. Statistical
dependence of adjacent message streams is ignored and independent message streams
originating at each node (terminal or repeater) are assumed; note that there are two such
streams at each repeater except repeaters 1 and N. S is measured in average number of
messages in the time taken to transmit one message (assumed constant). Thus S <1. Both
half-amplitude and zero capture were considered. When half-amplitude was analyzed, it wasassumed that the attenuation from repeater i + 1 to repeater i was less than that from
repeater i - 1 to repeater i. Hence, a transmission from repeater i+1 to repeater i cannot be
destroyed by a subsequent transmission from repeater i - 1. However, a transmission from
repeater i - 1 to repeater i will be destroyed by a subsequent overlapping transmission from
repeater i + 1.
These results were obtained by applying models based on work primarily by Tobagi and
Kleinrock [ 2 ], as well as by developing new models of our own. The initial results, which
are summarized in Table 5.1, are referred to as asymptotic because a large number oftandem repeaters are assumed. First, ALOHA is shown to have a greater throughput
capacity (0.5 versus 0.4) than CSMA with perfect scheduling, i.e., no collisions. ALOHA is
superior because there are times when CSMA will reschedule a transmission when it is not
necessary. Although these throughputs cannot be realized, they do establish upper bounds onobtainable throughput. These results are derived in Section 5.2. Also, as discussed in
Section 5.3, under full capture with CSMA, throughput is estimated as 0.25, but this is
regarded as optimistic.
The remainder of the results compare slotted ALOHA, unslotted ALOHA, and CSMAwith zero capture and half-amplitude capture. The CSMA results are based on results of
Tobagi and Kleinroek [ 2 1 and are derived in Section 5.3; the ALOHA results are derived in
Section 5.4. With either half-amplitude or no capture, slotted ALOHA is uniformly the best
54
nlacm
1'2 N-1i
FIGURE 5.1: TANDEM REPEATER (ONLY ADJACENT UNITS CAN COMMUNICATE)
nlac.TABLE 5.1: ASYMPTOTIC CAPACITY OF TANDEM REPEATER NETWORKS
ACCESS SCHEME CAPACITY
PERFECT SCHEDULING -ALOHA 0.5
PERFECT SCHEDULING -CSMA
CSMA - "FULL" CAPTURE 0.25
SLOTTED ALOHA -1/2 AMPL. CAPTURE 0.180
CSMA - 1/2 AMPL. CAPTURE 0.168
SLOTTED ALOHA - 0.15NO CAPTURE
CSMA - NO CAPTURE 0.137
UNSLOTTED ALOHA - 0.0881/2 AMPL. CAPTURE
UNSLOTTED ALOHA - 0.078NO CAPTURE
s.I.
IIac_technique, while unslotted ALOHA is uniformly the worst. Maximum throughputs of 15 to
20% are expected. Since the synchronization required for slotted ALOHA may not readily
be implementable in a network environment, it would appear CSMA is the superior
alternative. However, the CSMA results are based on results derived by Tobagi and
Kleinrock and include some optimistic assumptions.
In Section 5.5, a more detailed analysis is performed, which eliminates the asymptotic
assumption. The exact analysis of a three-repeater network analysis operating with CSMA
and half-amplitude indicates a maximum throughput of 0.268 versus 0.365 predicted by
results based on the Tobagi and Kleinrock model. From this we conclude that the CSMA
throughput estimates are optimistic and that the superiority of CSMA over ALOHA in a
network environment is marginal.
Then in Section 5.6, a lower bound is derived for the throughput that the C2 network
can support. The bound indicates that with high confidence, the network can support the
offered load in the scenario.
5.2 Perfect Scheduling Analysis
First we investigate the best that can be done if perfect scheduling is possible.
Consider the chain in Figure 5.2(a), where specific streams have been labelled with the
letters A to G. We schedule the A messages in an interval (average) of length S. The B
messages, to be successful, cannot overlap with A, and so they require a second interval of
length S. Similarly, C and D require additional intervals, as depicted in Figure 5.2(b).
Consider the ALOHA access method. E cannot be scheduled with A because node 2
can hear both. But E can be scheduled simultaneously with B, although nodes 2 and 3 hear
both transmissions, only nodes 1 and 4 can receive the messages. Similarly, F and A can
transmit simultaneously, and so on. Thus we have 4S< 1 or S <1/4. Defining capacity, C,
(per node) as a maximum value of 2S, we have C = 0.5 for perfect scheduling in ALOHA.
This is true if we use slotted or unslotted ALOHA, whether or not we allow half-amplitude
capture.
Next, consider CSMA. We assume that the propagation time between neighbors is
small enough to be negligible, i.e., that neighbors instantaneously hear each others' trans-
missions. The arguments proceed as above for A, C, D, and F and are illustrated in
Figure 5.2(c).
0 :4
nlacm
1 B 2 D 3 F 4 H 5
FIGURE 5.2(a): TANDEM4 REPEATER LABELED INFORMATION FLOW
I I II ~ 'TIME
A B C DF E H G
FIGURE 5.2(b):- POSSIBLE ALOHA SCHEDULING STRATEGY
A B C D EF G H I J
FIGURE 5.2(c): POSSIBLE CS!IA SCHEDULING STRATEGY
5.
rla ,But here, E and B cannot be transmitted simultaneously because of CSMA. (A more
elaborate scheme could be used to allow this.) But G and B and H and C can be transmitted
together. Thus, 5S <"S 1, S <S 0.2, and C = 0.4 for perfect scheduling in CSMA. This is less
than with ALOHA because of the times CSMA prevents transmission, even though it could
have been successfil (E and B).
5.3 Tandem Repeaters Using CSMA
In this section the hidden terminal analysis of Tobagi and Kleinrock [2] is extended
to the tandem repeater network using CSMA. Because of the hidden-terminals effect, this
analysis is approximate and is believed to give results which are optimistic, i.e., resultant
capacities are too high; more refined analyses are presented in Section 5.5. Consider a
group of terminals, all of which can hear each other, so there are no hidden terminals. As a
group they wish to send messages at a rate S to a station which is also in hearing range. We
model this source of messages as a Poisson process. Since each terminal can hear each
other terminal simultaneously, they will not transmit until the channel is idle. There will be
no collisions (assuming instantaneous propagation). If a message is not transmitted, it is
rescheduled at a random later time in the "far" future. The total stream of scheduled
messages (actual transmissions plus blocked transmissions) can also be modeled as a Poisson
stream of rate G >S. Consider the channel occupancy. A message is transmitted
successfully for 1 time unit (normalized). Then, the channel is idle for an exponential time
of average length 1/G. Thus, the probability that the channel is idle is
IG= 11+1/G 1+G
and is exact for this model. A scheduled message is transmitted successfully if the channelS= 1 Gis idle. Thus, the probability of success is a -+T or S =-1"-' and the maximum is
5 = 1. Obviously, this can be achieved only in the limit as G becomes infinite. Thus, the
success probability S/G goes to zero Pnd delays become infinite.
The existence of hidden terminals will lead to collisions. Tobagi and Kleinrock thenassume terminals are grouped according to their connectivity, i.e., all members of a group
can hear each other and the same other terminals. A terminal will transmit if its group is
idle and Its neighboring groups are idle. Otherwise, it will reschedule the message.
0 5.6
11iac_
Consider the effect of the neighboring groups on this decision. Let G' be the rate of
scheduled messages that are not blocked by the neighbors; G is the rate of scheduled
transmissions. Then (approximately), G' plays the role of G above. They assume that the1probability of being idle is +-,. Furthermore, they assume (optimistically) that each group
is independently idle. Then, if Ni is the set of neighbors of group i,
G.G I I (G ')
Je N i
Now consider the transmission of a message from group i to the station. A scheduled
message will be transmitted by group i successfully to a station if it is transmitted at all,
i.e., its group and its neighbors are idle, and if no other groups (hidden from the transmitting
group) transmit during the transmission time. These probabilities can be immediately
evaluated as:
1
I+G i Group i is quiet when transmission is initiated
kEN i 1+G7) - Its neighbors are quiet
kJK ( l+GkI - The hidden terminals are quiet where
K = Ni1 J{il.
11 -G' k
kfK e - The hidden terminals do not begin transmitting after
group i has initiated transmission.
Then, assuming independence, the probability of success is:
TI (l4G')Gi I (I+G'j
all j
5.7
PACKET RADIO DEPLOYMENT SYUOY.CU)APR 80 OAAKO7-C-0763
UNCLASSIFIED FR.207.01R1L
23l- 8 f1 EW NL YSISCORP G f l f l f l flNYF /6 17/2.
EEEEEEEEEEEE
""1.511111 4.
MfCR0e,'.0'Y HI M)I UION ISI COARI
-Gs In the above, the probability that a packet radio does not transmit is approximated by
e . Thus, the transmitted stream is also assumed to be Poisson.
For our problem, there are no stations; each group essentially consists of a single
terminal. Hence, the above analysis does not directly apply. In our problem, after therouting is distributed, the rate of transmission S between neighboring terminals can be
determined. We will assume that all these message streams are independent and Poisson.Furthermore, let Dici(NjA N) be the terminals which dominate a transmission from I to J,in the sense of half-amplitude capture, where the bar denotes set complement. If dij is thedistance between i and J, and if proximity determines capture, then kc D if kc Nil k t Nitand dkj Idij. Also assume separate (Poisson) streams for scheduled and unblocked messages
at each node for each destination. Then,
SiJ kDi e-G 'k
: Gi H (I+G )i£ NjUN i
-sjEG.
8i =
j CN
t 1Gi = Gi
The above expression is similar to the previous analysis. However, in this cse thenumerator is defined noting that only devices that have energy greater than group I canInterfere with a transmission from I to J; otherwise, they must remain silent. Also, thedenominator Is defined noting that terminals which I and j cannot hear are irrelevant (and
5.8
,nac _hence do not appear in the equation). The above approximate analysis can be applied to any
topology. Consider the topology of Figure 5.1, where we assume that terminals to the left
dominate. Using the above equations, we obtain after significant algebraic simplification:
G'Ii (+Gi_ 2) e G i- 2 + (I+G'i+2), 3<i<N-2
I+G'i
G = S (1 +0G3), .. 2 = S (2 + G4
1+G i I 2
n-i = S U+ Gjn 3) en3 + I] =e
l+G'n1 1+G' n
Assume that as N-w , we reach a steady-state condition where G'i - G'. (We
observed this behavior in an exact analysis of an ALOHA scheme). Setting all G=i G, we
have
= 0'8 GI
(l+G,)2 (l+eG ')
which achieves a maximum of 0.084 at G' = 0.5. Thus, C : 0.168 for CSMA with half-
amplitude capture. Again note that this analysis is approximate and probably optimistic.
The actual capacity will be less.
It is easy to determine the effect of half-amplitude capture with this analysis. TheC. above equation can be rederived. Let GI (G2) be the scheduled stream to the right (left) for
a typical terminal. Then,
1. S e-GO' 001 - 4 ' - ') 0=01 + G2 , 1=(1+0' 02 (1+0')
* 5.9
For no capture, the first term becomes
S *'<' and S= 1/2 G'e-G'= (1+G,) 4 (1+G,) 2
which has a maximum of 0.0684 at G' 0.4. Thus, without capture, C = 0.137. Similarly,
if somehow we can achieve "full" capture, the second term becomes
_ 1 and S = 1/2 G,G_ (1+G,) (1+G,)2
which has a maximum of 0.125 at G' = 1. Thus, C = 0.25 if we can have "full" capture.
These results are summarized in Table 5.1.
5.4 Tandem Repeaters Using the ALOHA Access Method
In this section an exact analysis of an ALOHA-type access scheme is presented.
Consider first the ease of unslotted ALOHA with half-amplitude capture. We have modelled
each terminal's traffic as one or two Poisson streams. If a terminal was operated in "pure"
ALOHA, its transmissions would collide with themselves if they appear less than one
transmission time apart. In traditional ALOHA analyses, one usually modelled a group of
terminals that could not hear each other. If they could hear each other, then CSMA is
ideal. Here, there is only one terminal in a group, and such collisions are in essence suicidal
(also physically Impossible) and should not be allowed.
We now modify the access scheme at a terminal. This is essentially a revision of the
message-stream process. The original message stream is Poisson with a rate S. If any
messages should collide because they are too close together, they are rescheduled at a
random time in the far future. Any messages which are transmitted and collide with
messages from other terminals are also rescheduled, as above. The resultant transmitted
message stream has a rate G > S because of these collisions, and the stream is no longer
Q 5.10
lnac_Poisson. Consider channel occupancy again. A message transmission takes one time unit(normalized). The time to the next message is exponential (by the above assumptions) with,
say, mean /G*. The fraction of time that the repeater is transmitting is
1 G* 00 = TG7GT " 0 +I -+ 1--G
Thus, the repeater is idle with probability 1 - G and will not transmit in a unit transmissioninterval with probability e7M- G( ). The important aspect of this model is that every
terminal operates independently.Assuming half-amplitude capture as before, and letting G1 (G2) represent the stream
to the right (left), we have
S ( - G Gi+ l 1
G M ~1 )exp - 7i +1) I"1 Gi+2)
= (-G)Ixp G I exp-2 _T ( 1-2')
G -12 1.GI (-2) 2) e
G = Gil+ G12
3 1 N -2
with obvious changes for I = 1, 2, N- 1, N. Note that we assume in the above that a
terminal can decide to transmit while receiving, and thus force a collision. This was done topreserve independence between terminals. If all G, G when N ' , as above, then
5.11
s (-2 e'-2l(1"G)
14e-GI(l-G)
which achieves a maximum of 0.044 when G = 0.2. Hence, the capacity is 0.088. Withoutcapture, similar analysis yeilds
S 1/2S(1-G)2 e-2G/(1-G)
which achieves a maximum of 0.039 at G = 0.18; hence, C = 0.078.In a similar fashion, a slotted ALOHA sevheme can be analyzed. Again assume that if
a terminal has already scheduled a transmission in a slot, it will reschedule any othermessages that fall in that slot. Then, terminals again act independently. Let P be theprobability that a terminal will transmit in a slot. Assuming that capture can take placerapidly, we again consider half-amplitude capture as discussed above. I - P is now theprobability that a terminal is idle. For N -'o
SPI (1- P)
S (1 -P)2P2
P P1 + P9
P S + S
1P (1-P)
Thus, where P1 is the probability that repeater I is transmitting to repeater i + 1, and P2 isthe probability that repeater i Is transmitting to repeater i -1,
P(1-P)2
which achieves a maximum of 0.0902 at P = 0.38. Hence, C -0.18. Without capture,
Sj 2
S8 1/2 P(1 -P) 2
and achieves a maximum of S = 2/27 at P : 1/3; hence, C : 0.15.
From these preliminary analyses, it is seen that slotted ALOHA, properly modified, isslightly better than CSMA, and that unslotted ALOHA has approximately half the capacity.
Other access schemes and more refined analyses are possible. However, it should be notedthat if echo aeknolwedgments were to be included in the analysis, the performance of the
ALOHA schemes would degrade more than that of CSMA.
5.5 Tandom Network Using CSMA (Exact Methods)
This analysis is more refined in that the results do not require the number of repeaters
to become arbitrarily large.
5.5.1 Three Repeater Problem
To get a better feel for the approximations in the approach based upon Tobagi and
Kleinrock'a work, we have developed alternate analytic techniques for this project. Firstconsider a small chain with N = 3. Let PI equal the probability that the entire system is
idle, and consider CSMA with half-amplitude capture as defined above. Then,
S 2S S -G1=PI = -,- = P, e
1 02' 3
where Gi are the Poisson rates of the scheduled traffic, 1 dominates 3, and eG1 is the
probability that 1 does not collide with 3. Let Ik be the event that the kth repeater is idle
at an instant, k = 1,2,3. Since packets from repeater 2 never collide, P(I2) : 1-2S. Also,
PI P(11 2 3)= P(x x31 2) P(12)
It can be shown that
P011) 1) 0 P112) P(I3 I1)
323
The argument is essentially that when repeater 2 is idle, the other repeaters are operating
independently. It can also be shown by similar arguments that
PO 11 1 +G2 = 1 + P31 1+G3
which is the same expression as if each repeater were acting alone. From the above results,
S can be determined as
= 11S (1+G X(G1 G
1)(+G 1e 1) + 2G 1
This expression has a maximum of 0.134 when G : 0.5. Thus, C = 0.268. This is an exact
result. From the techniques based upon Tobagi and Kleinrock's methods, we obtained a
capacity of C = 0.365 for the same problem. Thus, it is believed that their results are
optimistic. Note that we apply the results of Tobagi and Kleinrock in a fashion probably not
anticipated by them, and the above comments are not intended as criticism of their work.
These results can be generalized to a star configuration where an arbitrary number of
repeaters all home on a single site as depicted in Figure 5.3. In this case
PQ = 1 - 2nS
where
PRU one is the hub
- There are n surrounding PRU's.
Similar independence results can be established showing
P 1iI 1 ) = PUiII) P(IjlI1 ) i J i1,jIA1
Pa(I) 1/1+G1 .
Q 5.14
2 5
S S
FIGURE 5.3: STAR NETWORK (n *4)
- ---------------....
_S e (n-1)Gi
G1 InS
under the assumption of no capture.
5.5.2 Multiple Tandem Repeater Analysis
We assume that routing has been accomplished and the packets scheduled fortransmission from a node to its neighbor are represented by a Poisson stream, which is
independent of all other such streams, i.e., going to another neighbor or at another node.
These scheduled packets may or may not be transmitted. If when a packet is scheduled, thetransmitting node and all its neighbors are idle, then the packet will be transmitted. If not,
the packet will be rescheduled at a random time in the future. A transmitted packet will be
successfully received If In addition all of the neighbors of the receiving node are idle and
none of them begin transmission for the duration of the packet. (When half amplitude
capture is in effect, some of the neighbors may transmit without destroying reception.) If
packets are unsuccessfully transmitted, or cannot be transmitted, or if acknowledgments arenot heard, the packets must be rescheduled. We assume that all rescheduling is at a long
and random future time so that the entire scheduled stream of traffic is Poisson. Thus we
model the traffic as independent Poisson streams at each node. Let ni be a neighbor of node
i. Then G. is the rate of the scheduled traffic from i to n. All the streams, for every ii,n. I*and ni , are asumed independent. The total scheduled traffic at a node is also Poisson and
has a rate
Gi : nGi,ni
The performance of the network depends upon the status of the nodes when a packet is
scheduled. Let A denote a set of nodes and P(A) the probability that all nodes in A are idle.
If B is another set of nodes, then P(A,B) = P(AfB); and P(A I B) is the probability that all
nodes In A are Idle given that all nodes in B are idle. We can now provide some general
results.
0 5.16
let Ni be the set of all neighbors of a node i. Then for any AC.Ni,
P(i,A) = P() + P(A) - 1. (1)
To prove this result, note that P(A) = P(i,A) + P(6,A), where i denotes that i is busy (not
idle). Then P(i,A) = P(A i)P(i). However, P(A i) = 1 and P(i) = 1 - P(i).
A node i will transmit when i and all nodes in Ni are idle when a packet is scheduled.
If the length of a message (or its average length) is normalized to unity, then
P(i) = 1 - P(i) = GiP(i,Ni). (2)To understand this equation, note that Gi P(i, Ni) is the rate of the actual
transmissions because i will transmit if and only if i is idle and its neighbors N. are idle.
Consider a long time period T, then the fraction of time during T that i will be transmittingis
P( i G = i P(i, N) T/T.
The result follows immediately, noting the cancellation of T.
Now Gi is the scheduled packet rate in units of packets per (average) packet
transmission time. Letting A = Ni above in (1), combine with (2) and apply definition of
conditional probability, we get
P(iI Ni) = (3)I 1+Gi
Note this established the independence results in Section 5.4.
We consider first exponential packet lengths, again normalized to unity. The evolution
of the network is now completely characterized by a continuous time Markov chain. Let D
be a disjoint set of nodes, i.e., none of which are neighbors. LetObe the class of all sets D,
including 0, the null set. Let D be the event (state) that all nodes in D are busy and all nodes
not in D are idle. Then 9 denotes the event that all nodes are idle. Let N be the set of all
nodes in the network. These states form a Markov process. The transition equations are
(E G PO = E P() (4)\iCNi icN
5.17
For D 0, and d = D the cardinality of D, and f(D) the set of nodes that are disjoint
with all nodes in D,
(d + ZDGj )P(D) = E Gi P( CDT ) + Z P(17y j)i (5)j cfAD)/ ieD j cf(D)
4 It is relatively easy to see that these equations are solved by
P(D) =( G) P(a) (6)
6 P(O) is the probability that all nodes are idle and is given by
-1P(D) = [+ E - (7)
With these results, the probability of any occurrence in the network can be determined in
terms of the' { Gi }. For example, let the set of nodes and Q be the set of sets D containing
9 none of the nodes in A, including 0, then
0 P(A) = Z P(D) (8)Dc Q
These probabilities can now be related # the successful transmission rates from each
node. Let Si,n be the rate of successful transmissions from a node i to its neighbor ni .i~nThen
Si,n i
.-- = P(Ni,N )QiiG ini i 1(9)
Note that for a scheduled packet at i to be sent Ni and i must be idle and i c Nn* For this
packet to be received at ni initially then Nni and ni must be idle and n, c Ni. Then
Qi,ni - P(during a transmission time no neighbors of ni transmit thatare not dominated by i I Ni,Nni, and i is transmitting).
5.18
Inac_The probabilities Qi,n. can be computed for exponential packet lengths. The above equation
is valid for constant Oacket lengths but Qi,n is much harder to obtain, although bounds and
approximations can be found.
Example 1:
(Assume half amplitude capture with left nodes dominating)
S12 = P(1,2,3) =- S21
$231
_23 = P(1,2,3,4)
G23
$32- = P(1,2,3,4)Q 32
G3 2
= P(2,3,4)
G34
$S43= P(2,3,4)Q
G43
G1 = G1 2, G2 G21 + G23 , G3 = G3 2 + G3 4, G4 = 43
P(l293,4) =P(O) = [1 + G1 2 + + + G 1 G3 + G1 G4 + G2G41
(1,2,3) = P(W + P(4) = (I + G4) P()
5.19
n1ac.P(2,3,4) = P() + P(1) = (1 + G1) P(A
Here we have assumed half-amplitude capture with left nodes dominating.
Q32= P(1 doesn't transmit during a transmission time all nodes idle
initially and 3 is transmitting).
Since 3 is transmitting 2 must be idle and I will transmit if scheduled. Thus Q32 is the
probability that the packet ends before 1 transmits.
132 1+G
Similarly,
Q43= P (2 doesn't transmit during a transmission time 2,3,4 idle initially
and 4 is transmitting).
Since 4 is transmitting, 3 must be idle throughout. There will be a collision of 2 transmits
before the packet ends. Node 1 may or may not be idle. If 1 is not idle, then 2 cannot
transmit. Thus
1- 43= P(1 2,3,4) a + I - P(1 12,3,4)]
a- 2 + (1/2) a
1 1+G +G2 I+G +G2
= 1/2 M
Here a = P(2 transmits before packet ends j 1,2,3,3) and can occur if 2 transmits first or 1 ]'transmits first, ends before the packet ends, and then 2 transmits first. 1 is defined
similarly.
5.20
0
____ ___ ___ ____ ___ _- a c.8=P(2 transmits during the period 11,2,3,4)
The equations can not be solved for' {S w in terms otf {G. 1. Maxi mum values of
S.j can be found to determine capacity. The effects of ~ ieacknowledgments canalso be evaluated. This will relate the' (S I to the aetivil end-to-end successfultransmission of packets. Si wini then include repetitions. 74),, 's rre procedures can be
used for any network topology.
Example 2:
(Again assuming half-amplitude capture with left nodes dominating.)
S12S21 S23
- = P(1,2,3) P(O)-2321 I1+G1G 2G 3 +I*
Let S12 = I 12 = G 1 , S2 1 + S23 =S 2 , G2 1 G 23 G 2 1 ' 3 2 ' 3 , G32 = 3
Then
8 j 82 83 -
G1 -U2 3 U-( + G) = I+G Xl+G )+G2
For specifity, let S 1 = , s 2 =2S, S3 =S.
Then
G02 =2Gj, 03 G, U1( + Gi) and
5.21
Th axmmvau f ondt 1 1 G4 l
S G= 2 3
(1+Gl)[I+Gl(l+Ol)] +G, 1 4G, + 2G1 I1 G1
The maximum value of S found to be 0.138 for G1 = 0.6.
The above analysis assumes exponential lengths for the packets. There are some
unusual consequences of the exponential model however. We have seen that a similar
analysis of a three-node chain with constant packet lengths results in a throughput of 0.1348
which is slightly worse than that for the exponential assumption. The reason is that smaller
packets are more likely to be successfully received. But the throughput here, S, is in
packets/average transmission time. When the lengths of the successful packets are
considered, the exponential assumption results in a lower throughput, at least in the example
below.
The dependence of probability of success on packet length refers only to those
collisions due to the "capture" effect. For the three nodes chains above, assume 3 is
transmitting to 2. This will happen if 2 and 3 are idle. If I is also idle, 2 will begin to
receive. So far, the length of the transmission has not played a role. But there will be acollision if 1 schedules (and transmits) a packet before 3 finishes, and this does depend upon
packet length.
Let P32 be the probability of successfully sending a packet from 3 to 2. This was
$32/G32 above. Let P3 2(x) be the probability given that the transmitted packet has length
x. Then
-GxP32(x) = P() e
Averaging over all packet lengths we get
-xSp 3 2 (x) e dx = P(fD) 1 32
032 1+ = G32
as it should.
When a packet of length x is successfully received, then x bits (normalized) are
delivered. The successful "bit" rate (normalized) is given by
M -x SIx3 (x) e GP(O2Q32 f x = (P33 1+2 )2 1+G1
0
This compares to S32 for the bit rate when the packet lengths are constant. The value of Gthat gives the maximum throughput will be different however. For the exponential lengths
we get a bit rate of .096 when G1 = 0.3. Thus the throughput in bits is indeed less than
that for constant packet length.
There are other consequences of this phenomenon. The exponential packet lengths inthis model are independently regenerated for every transmission for the same message.
Thus eventually a small length will be tried and the probability of success enhanced.
5.6 Application: A Loose Bound on Performance
Consider a general network with a host, i.e., all traffic directed to or from a specific
node, operating under CSMA with half-amplitude capture and including the effect of passive
acknowledgments. We can replace this network with a simpler network with a much lower
throughput and thus derive a loose lower bound on the throughput.
The network operating as defined above (CSMA, etc.) should perform no worse than if
unslotted ALOHA was used. (If not, then CSMA should be replaced by unslotted ALOHA.)
Thus, we consider unslotted ALOHA for the network to derive the lower bound. We find a
set of minimum hop paths for the network. Again, we need not find the best routing since
we are looking for a lower bound. However, the bound will be tighter if a better routing is
found.
Consider two nodes both i hops away from the host. They may or may not hear each
other and may or may not hear the same set of neighboring nodes; note, a node at hop i, in a
minimum path route, can only hear other nodes at hops i - 1, i, and i + 1. These two nodes
may simultaneously receive or simultaneously transmit if they are corresponding with
disjoint neighbors. If we merge these nodes, i.e., replace them by a single node with theaggregate traffic and a set of neighbors equal to the union of the neighbors of each, then
under an unslotted ALOHA discipline, the performance cannot have been improved. (In
CSMA, however, the performance could have been improved because there will be less
collisions.) This step can be repeated until we are left with a chain of nodes with the host at
one end. The length of the chain is that of the longest path in the min hop routing. The
traffic to and from a node in the chain is the sum of the traffic of all nodes at the same hop
level in the min path routing.
This chain can be analyzed using unslotted ALOHA. For simplicity, we move all
traffic to the end nodes and consider, S, the larger of the two rates, one for each direction.
5.23
ilac_Thus the chain is modeled as attempting to send S packets/second from the host to the last
node in the chain and S packets/second in the reverse direction. S is the larger of the sums
of all traffic for all nodes in either direction. The performance is certainly not better than
if each node is permitted to have different requirements. This step need not be taken if the
resultant bound is too loose.
Consider the host which is attempting to send S packets/second. Because of collisions
with itself and the next two nodes (for the bound we consider zero-capture here) and
because of failure to hear passive acknowledgments, packets will have to be repeated. As a
worst case, we assume that all packets will be repeated exactly K times where K is the
maximum number of tries allowed in the protocol. However, the next node will notice
repetitions and attempt to transmit only one copy. Thus the rate of scheduled messages at
the host, G1 , which is in general bounded by KS, is here set to KS. Not all the messages will
be successfully received by the next node. Thus, its transmission rate will be less than S.
Again, assume the worst, set it equal to S, and the scheduled rate of intermediate nodes, in
each direction, is at most KS. The total scheduled rate at each intermediate node is thus
G = 2KS. For simplicity, we inflate the scheduled rate at the end nodes to 2KS also.
For all but the last node in a chai, there will be a collision if during a transmission
the sending node or the receiving node or the next node is already transmitting or begins to
transmit. These nodes thus cannot successfully transmit a packet in an interval of two
transmission times surrounding the start of transmission of the packet in question. We here
assume constant transmission times and normalize G and S to rates in packets/unit
transmission time. We also assume that all streams are Poisson and independent. Thus, the
probability of no transmission in the interval by any node is given by e-6G = e -2(2.KS) To
be successful, three successive nodes must be quiet, thus the probability of success ise -6G = -12KS P. For the last transmission in the chain, only two nodes need be
quiet, but again for simplicity we use the smaller P above.
Since a packet will be repeated at most K times, the probability of successfully
transmitting a packet over a link is at least 1 - (1 - p)K. The probability of transmitting a
packet successfully over £ hops is q = 1 -(1 - P) where P = e 12KS.
To apply these results, it is first necessary to compute the network traffic volume
during the peak hour in units of bps. The traffic rate (including overhead) can be determined
as follows
T M IP + n(Poh) + n(ETE)]3600
5.24
where
T = Traffic volume in bps during peak hour.
M = Number of messages during peak hour.
IP = Text length of message.
n = Number of packets per message (3).
Poh = Packet overhead (448 bits).
ETE = End-to-end acknowledgment size (448).
These traffic calculations are summarized in Table 5.2. As indicated in the table, the peak
hour rate from PRU's to the host is 1065 bps while the peak hour rate from the host to the
PRUs is 1035 bps. Since the first value is larger, it is used as the throughput constraint in
the above model. Thus with a 100 Kbps data rate, the network must support a throughput of
S = 0.01065.
For a 3 hop network with 3 transmissions,
-12KSP =e
-12(3X.01065)P =
P = .698
Then
q = -(i.~K]P
* .1 3I- U..98)
- .92
Thus with confidence greater than 0.92, a three-hop network can support a throughput
S = .01065.
.. . ... .. . ". . .. . . .( .. . I. .. ~ i"
... -. ... . . .. . . . . . . - - , . . . .. .. . . . . .. .
n1ac.TABLE 5.2: C2 TRAFFIC VOLUME HOST SCENARIO
PRU-0.HOST HOST-4-PRU
Transmit Messages 4692 4692During 30 Hours
Direct Messages 2742 2534During 30 Hours
Total Messages 7434 7226During 30 Hours
Total Messages 55578During Busy Hour
Information PacketOverhead Bits/Packet 448 448
ETE ACK4844(Bits/ACK)4848
Message Size 3760 36(Information Bits) 36
Bit Rate 1065 1035(bps)
C
Aft
nac__6. SURVIVABILITY MODELS
In the initial analysis of network conectivity using the Longley-Riee meanattenuation model, the PR nets were sparsely connected. At this time, it appeared that
survivability was a major problem. Thus the analysis described below was performed to
determine how many repeaters would have to be introduced to guarantee a certain level of
survivability. However, with both the random Longley-Riee and LOS link models, there is
substantially more connectivity and survivability is not a major problem.
Typically, a minimal number of repeaters are required, and their selection can be
"eyeballed." For example, in the C2 subnetwork (initial deployment), two repeaters were
selected by MITRE with locations
* - Repeater 1: 500 210
- Repeater 2: 590 394.
The critical devices in the C2 network are the Maneuver Battalions. Without these
repeaters, the Maneuver Battalions 1-78, 1-81, and 1-79 were essentially isolated having
only one link connecting them with the rest of the network. Clearly, this would be an
unacceptable level of survivability. However, with the introduction of dedicated repeaters,
each of these devices could communicate with the rest of the network via 3 different paths.
Thus, they would be isolated only if 3 devices failed. MITRE has estimated that the survival
probability of all devices is greater than 0.95; thus, the probability of a Maneuver Battalion
being able to communicate with the rest of the network is greater than
91 - (I-p)3 = .9998
for P = .95. Thus, we could conclude survivability is not a major problem. The models
C described below were developed when it was believed that the network connectivity was
sparse. We have examined three models for insuring a specified survivability level. In all
cases specified, survivability levels can be achieved by appropriate placement of nodes. In
the following, we assume nodes fail independently with a uniform rate p. We consider the
C probability of two nodes, s and d6 separated by paths with h intermediate nodes being able to
communicate.
0 64
n1ac.6.1 Models
6.1.1 Model A: Independent Paths
In this model (see Figure 6.1), we assume that all paths from s to d are independent. In
particular, we assume that a node on one path cannot communicate with any node on
another path. Thus, communication from s to d can only take place if an entire path
remains intact. The probability of a single path working is sh = (1 - p)h. Let s = the
probability of at least one of r such paths working is 1 - (1 - sh)r. If sh << 1 and r >> 1,
this probability is approximately I - e-rSh. Thus, given a desired survivability, R, and
values for p and h (hence sh), we find r is given approximately by:
s h (1-P)h
Thus, r grows exponentially with h and if p is significant (say 0.2), r grows very quickly, as
shown in Table 6.1 which gives exact values of r required for survivability of .95 and fortypical values of p and h. We conclude that independent paths do not provide a sufficient
level of survivability for reasonable values of r; some form of overlap among paths is
required. The model of independent paths is useful, however, as it is a worst case for a
given value of h and a given number of nodes.
6.1.2 Model B: Fully Connected Groups
This model represents the opposite extreme from the previous one. For most
* situations, the best one can do given that one can place only hr nodes, all nodes fail
independently with probability p, and all paths must have h or more intermediate nodes in
cascade, is (see Figure 6.2) to place the nodes in h groups of r such that all nodes In each
group can communicate with one another and with all nodes in the preeeeding andsucceeding groups. The probability of all nodes in a group of r nodes all falling is pr. Thus
the probability of s and d being able to communicate is (1 - pr)h. If hpr<< 1, this is
approximately equal to 1 - hpr. Thus, given a desired and values for p and h, we find
64
nacmr INDEPENDENT PATHS
S
h INTERMEDIATE NODES
v d
FIGUE 61: NDEENDET PTHS(LIKS IDICTE AIROFNOE WICCA IRETYCMUIA
6.3
TABLE 6.1: PATH REQUIREMENTS (r) FOR INDEPENDENT PATH MODEL
(Survivability Level .95)
p =.05 p .2p 5
h 2 2 2 3 11
h 4 2 3 6 47
h=6 3 4 10 191
h 8 3 6 17 766
h 10 4 7 27 3067
C
0GA4
nac..
S
I' 'II" I~I II I II I II II II I iI I II I II I II ,~I, 'II,
1~'I
4
FIGURE 6.2: FULLY COIIUNICATING GROUPS
4
Q 6.5
nac._log 2 P
Thus, r grows very slowly as a function of h. Table 6.2 gives (exact) values of r for typicalvalues of p and h for R = .95.
6.1.3 Model C: Overlapping Chains
Fimlly, we consider (see Figure 6.3) a model which is neither an upper nor a lowerbound on reliability, but rather, a representative model of a chain where non-adjacent nodescan communicate. In particular, we assume that each node can hear the r nodesimmediately preceeding it and the r nodes immediately succeeding it in the chain. Thus sand d can communicate if no set of r adjacent node fail. Let c(h,r,p) be the probability thattwo nodes can communicate given there are h nodes in the chain separating them, nodes cancommunicate with r preceeding and r succeeding nodes, and nodes fail independently with
probability p.
Then we can write the recurrence relation:
r(I - p) pi- e(h - i,rp) for h > r
c(hr,p) = 1=11 for h >r
where the ith term corresponds to node i working and nodes 1,...i-1 failing. (It is assumedthat nodes s and d work.)
The general form of the recurrence relation is difficult to solve. Table 6.3 givesvalues of r for a survivability level of .95 and typical values of p and h.
6.2 Conclusions
There are two ways of obtaining the desired survivability given a number of nodes anda range over which communication can take place:
- add more nodes
6.6
TABLE 6.2: GROUP SIZE (r) FOR FULLY-CONNECTED GROUPS
(Survivability R =.95)
P =.05 P =.1 P =.2 P .5
h=2 2 2 3 6
h h4 2 2 3 7
h 6 2 3 3 7
h 8 2 3 4 8
h 10 2 3 4 8
2
3 h
ir
hd -
FIUR 63:-OELAPIG HAN
nlac.
TABLE 6.3: REPEATER RANGE (r FOR OVERLAPPING CHAINS
(Survivability Level =.95)
P =105 P =.1 P =.2 P =.5
h=2 2 2 2 2
h 4 2 2 3 4
h=6 2 2 3 5
h=8 2 3 3 3
h 10 2 3 3 6
c
c
- increase the range.
We explore the effect of each of these using all three models, for R = .95.
Suppose s and d are 10 Km. apart and the range is 1 Km.
Case A: P = .05
Model A: 36 nodes required
(4 chains of 9 nodes 1 Km apart)
Model B: 18 nodes required
(9 groups of 2 nodes. Groups 1 Km apart)
Model C: 19 nodes required
(chain of 19 nodes .5 Km apart)
Case B: P = .2
Model A: 189 nodes required
Model B: 36 nodes required
Model C: 39 nodes required
Case C: P = .5
Model A: 13,797 nodes required
Model B: 72 nodes required
Model C: 99 nodes required
Thus, we see that for most situations of interest, Models B and C give very similar results.
The importance of this is that nearly the same survivability is obtained by employing a
practical strategy (Model C) as can be obtained by an optimal strategy (Model B).
6.10
7. ROUTING
7.1 Overview
For the scenarios using LOS links and the scenario using direct-terminal
communications, typically source and destination PRU were separated by at most two hops.
In these cases, there is substantial connectivity and shortest arbitrary path routing is
adequate.
However, in the host scenario using links determined by the random Longley-Rice
model, more hops were required to traverse the network from origination to destination.
For this case, a heuristic routing algorithm was developed. This algorithm, which is
described below, provides shortest path routes but selects the shortest path in a way to
minimize congestion. The computational complexity of this procedure is minimal and could
easily be implemented in a PR station. Since only a few runs were made for this scenario,
the algorithm was not implemented on the computer. Instead, the algorithm was executed
manually and the resulting routes were entered into the computer. This exemplifies the
minimal complexity of the procedure.
7.2 Motivation
The unusual nature of the network connectivity (as determined by the random Longley-
Rice algorithm), in particular its relatively weak dependence upon distance, which arises
from the random connectivity model makes the routing procedure described in the following
sections necessary. The procedure is based on the following procedure which is suitable for
gt networks with sparse traffic matrices and distance dependent connectivity.
Consider the network shown in Figure 7.1, where node G is the host. The S1 are
sources of major traffic flows. All nodes are assumed to have some traffic requirements
and all communication takes place through the host, G. The goal is to produce routes which
C. have good throughput versus delay characteristics. This gives rise to several conflicting
objectives. Min-hop routes minimize the number of times traffic is repeated and hence
reduce the total traffic "in the air" as well as delay (for a given hop delay). Routes that
separate major traffic requirements onto paths with no overlap except at G will tend to
7.10 :
\PR W mm!'m ! i - / I acm
MAJOR ROUTES
K - OTHER ROUTESOTHER CONNECTIVITY
FIGURE 7.1: NETWORK WITH DISTANCE DEPENDENT CONNECTIVITY AND
SPARSE TRAFFIC MATRIX
UQ
7o2
improve throughput by reducing congestion in any local area. Routes which form chains,
i.e., in which no three nodes on the route can hear one another, will reduce congestion and
interference.
Routes which concentrate traffic to G through a small number of nodes will tend to
reduce interference at G since the CSMA is most effective when there are few hidden
terminals and separate routes into G (as we described them) would be hidden from one
another. The major routes shown in Figure 7.1 as heavy solid lines represent a compromise
among these objectives; these routes are not necessarily optimal. They are min-hop and
minimally interfering. There are three of them, and they are essentially chains. The
existence of such non-interfering, chainlike min-hop routes is a consequence of the distance-
dependent nature of the connectivity.
7.3 Routing Procedure
We now describe the proposed routing procedure in detail. The procedure is based
upon the principles outlined in the preceeding section and adapted for use in networks with
random connectivity. The procedure produces routes with the following characteristics and
objectives:
- The routes are min-hop.
Two nodes one hop (Level I nodes) from G are identified. Ideally, these nodes
cannot hear one another nor can they hear any other node one hop from G.
- Ideally, all traffic is divided evenly between these two nodes and enters G
U- through them.
- The two major routes (one through each of these nodes), ideally, are chains and
totally disjoint.
The worst case, using this procedure, would be that no two disjoint nodes could be
found or that essentially all of the traffic entered through one node. Previous analyses of
chains showed that the difference in performance is a factor of two in these two cases.
Thus, we have bounded the effect of the procedure.
7.3
nac.o
The procedure described is heuristic for many reasons. First, the network is feasible
even in the worst case, as has been shown in the section on throughput analysis. Second, a
rigorous treatment of the problem is likely to be time consuming eomputationally since even
the problem of partitioning the nodes into two groups of equal size, in the absence of any
other constraints, is NP complete. Finally, the random and unpredictable nature of the
connectivity makes execution of the procedure after the actual deployment a necessity and
hence, the ability to use a procedure which can be executed quickly with a limited amount
of computing power seems highly desirable.
The procedure works in two phases. Phase I identifies likely candidate Level I (the
level of a node indicates the number of hops from the host) node pairs to be evaluated in
Phase U. For small networks, it may be desirable to evaluate all disjoint pairs since the
quality of the solution seems to depend more on the solution of the node pair than upon
anything else.
Phase I: Host Routing (Identify likely Level I pairs)
Identify all nodes at Levels I and H by min-hop labeling from the host (level of node
equals number of hops that node is from host). Consider all pairs (;,j) of Level I nodes which
cannot hear each other directly. (Possibly, restrict (i,j) to those independent of other
Level I nodes as well).
For each such I and j:
Let S. = " K IK cLevel H and i can hear K directly}
S = defined analogously.
Compute a figure of merit, f(i,j), for the pair as follows:
f(i,j) = mfn(I Si - SjI , IS - Si 1 )
Do Phase !I for one (or more or all) pairs (i,j), considering those with the largest f(i,j) if
only a limited number of pairs are to be considered. If only Phase II is to be done for only a
single pair (I, J), choose the pair with the largest f(i, J).
7.4(I
1'iac_Note the selection of the f(i,j) is somewhat arbwtrary; for example another criteria is
minimum { Si$j
Phase 11 (Divide traffic between i and j from Phase I)
1. All Level I nodes are assigned to the host.
2. Assign all Level 11 nodes which can be heard only by I (or j) to i (or j).
2. Starting with the Level H node with the largest traffic, assign nodes which canbe heard by both i and j to i or J, whichever is more lightly loaded. Unless theseLevel 11 nodes already account for a majority of the overall, traffic, this "load"
should simply be the number of nodes.
4. All other Level 11 nodes, i.e., those that cannot be heard by i or j, starting with
the one with the largest traffic should be assigned to the Level I node with thesmallest load (traffic).
At this point, all Level I and 11 nodes are assigned (and have routes).
5. Any node reachable only through paths heard only by i (or j) should be assigned to
one of the paths. We work outward by level and descend in traffic magnitude.
6. Nodes reachable through 2 paths, one heard by i or j and the other heard by j only
should be assigned to the node with the smaller load.
7. Nodes reachable through paths heard by both i and j (i.e., paths containing a
Level 11 node hear by both i and j) should be assigned to the path with the morelightly loaded Level I node.
8. All other nodes (not reachable through any Level H node assigned to i or J) shouldbe assigned to a Level 11 node (reachable) which is assigned to the Level I node
with the largest traffic. (TIes broken by largest traffic at Level 11, etc.)
7.5
With this procedure, most nodes and most traffic should be assigned through i and J.
7.4 Computational Experience
The procedure was carried out manually on a version of the C2 network with random
connectivity at 150 dB. The assumed connectivity is given in Table 7.1. The traffic was
assumed uniform. A typical solution is given in Figure 7.2. As can be seen, the procedure is
reasonably effective, partitioning the traffic fairly well. The fact that it was possible to
carry out the procedure manually (in a few minutes) is indicative of its low computational
complexity.
7.6
iiocLLAU,co
ulzC~~~. -c wL
r- ULJ
z a.- am
S U.
to L&A
U, U.L) I-J
C-
2c
cn co o :E
040
Ln0
co-
CCY
1acI_8. REPEATER LOCATION ALGORITHM
8.1 Introduction
As discussed in Section 2, the configuration problem in its most general form isintended to select an optimal set of repeater locations and routes for all requirements.These requirements include a set of terminal locations, point-to-point traffic requirements,
a network performance model (which relates traffic flow to delay), a delay requirement, acoverage requirement, and a set of potential repeater locations (including all of theterminals). The terminals may move, and several snapshots are taken of their locations.Hence, the solution must satisfy all snapshots. Routing patterns and requirements may
change from one shapshot to another, but repeater locations (other than terminals acting as
repeaters) may not change.
This analysis considers a simpler problem focusing on connectivity issues. We are
given a set of terminal locations, T ='{ t. I i = 1,...,N } , a set of potential repeaterlocations, R = r i I = 1,...,M I, and a coverage requirement that each terminal be able toreach at least K of the selected repeaters. Under the assumption that the repeaters are
densely interconnected, this would imply that there are K independent paths betweenterminals. However, providing a K covering is not sufficient to guarantee K independentpaths, but as discussed in NAC's Third Semi-Annual Report, in most cases K independent
paths will result. A coverage matrix, C = Cij I i = 1,..., N; j = I,..., M 1, indicating whichterminals can reach which repeaters is defined such that Cij is I if ti can reach r., and iszero otherwise. Thus, in this analysis, we ignore issues of routing, link capacity, details ofthe link model, and delay and multiple time periods. We focus on the problem of covering
the terminals with repeaters.
In our case, the set of potential repeater locations is the set of terminal locations plus
some additional fixed repeater locations. It is most desirable to use repeaters which areterminals in the same subnet as the terminals to be covered. It is somewhat less desirableto use repeaters corresponding to terminals in another subnet, and least desirable of all touse fixed repeater locations. The repeater costs (weights) reflect this fact. The
methodology developed below is applicable for any of these cases.
A procedure is described below for solving the Set Covering problem (for K = 1) basedon a branch and bround algorithm. This procedure was developed before the LOS link
; 8.1
criteria was adopted. With the adoption of the LOS link criteria, there is substantial
connectivity in the network. Hence, the repeater location problem is essentially academic.
Hence, this procedure was not implemented as part of this contract. However, this
procedure was implemented for other NAC contracts and has demonstrzted marked
efficiency in solving set covering problems.
8.2 Set Covering Problem
The algorithm proposed for solving the repeater location problem is based on a branch
and bound procedure as illustrated in the flow chart (Figure 8.1). The algorithm consists of
the following basic steps:
Dominance check (Step 1): A one-time operation to detect redundancy
and reduce the size of the problem.
Branching (Step 2): Defining subproblems with specific repeaters
included and excluded from the solution.
Bounding (Steps 5 and 8): Computing upper and lower bounds corres-
ponding to a subproblem.
Fathoming (Step 6): Determining whether a subproblem has
potential for leading to an optimal solution.
Each of the steps are described in more detail below.
8.2.1 Dominance
Certain repeaters and terminals are redundant and can be eliminated from
consideration in the set covering analysis. The following observation enables us to detect
redundancy:
Observation 1: If (Cj = 1):= (Ck = 1), then column j can be eliminated from the
problem.
8.2
iiacm¥ ..
~1~
DOII NANCECHECK
BRANCH:DEFINE TWO SUBPROBLEMS
BY "FORCING" BEST CANDIDATE"IN" AND "OUT" OF SOLUTION;
INSERT INTO STACK
3.POP A SUBPROBLEM
FROi STACK
YES CHOOSE SOLUTIONLEFT CORRESPONDING TO STOP
?? CURRENT UPPERBOUND
5.SOLVE RELAXED SUBPROBLEMOBTAINING LOWER BOUND, LB
6.SYES FAHM
COMPUTE UPPER BOUND, U,FOR CURRENT SUBPROBLEM
r. COMPUTE NlEW UPPER BOUND
un UB MINIM'U'M (U, UB)
FIGURE 8.1: SET COVERING ALGORITHM
8.3
n1ac.Observation 2: If (Cij = 1)=:(Cj = 1), then row I can be eliminated from the
problem.
These observations are, respectively, that repeater k covers all terminals repeater jdoes, and that terminal I is covered by every repeater which covers terminal i. Clearly, anycovering including repeater j could be replaced by a covering using repeater k without
degrading the solution. Thus, repeater k dominates repeater j. Similarly, any covering
which covers terminal i will also cover terminal 1. Thus, terminal i dominates terminal .These dominance relations can be interacted, i.e., after eliminating dominated repeaters and
terminals based on the original covering matrix, one may eliminate further repeaters andterminals based on the reduced matrix. In practice, this technique has proven to be very
effective, substantially reducing the size of all sample problems (tried by hand with 20 to 50
terminals and repeaters).
8.2.2 Branching
The branching procedure for this algorithm traverses a binary tree as depicted inFigure 8.2. To begin one repeater is selected, say R. ; then in all solutions descending fromthe left branch in the tree Ri is included ("forced in) while in all solutions descending from
the right branch Ri is excludid ("forced out"). Analogously preceeding to the second level,
another repeater is "forced in" on the left branch and "forced out" on the right branch. Thisprocedure is then iterated until a node is defined for every possible combination of
repeaters.
The criteria used for selecting a new repeater to be forced in or forced out is based onthe number of terminals covered by the repeater. The repeater which
- has not yet been "forced," and
- covers the most terminals (that are not already covered by repeaters "forced in"
in the current subproblem),
is selected as the repeater to be "forced" in the next level. This is strictly a heuristic
criteria and could be replaced by any other arbitrary criteria.
8.4L
nac
R1 R.j
1 2 R1 Ri2 RjR 1 IRj2
CONVENTION: BAR OVER REPEATER NUMBER MEANS REPEATER FORCED OUT
NO BAR OVER REPEATER NUMBER MEANS REPEATER FORCED IN
FIGURE 8.2: BRANCH AND BOUND TREE
* N . .
8.2.3 Fathoming
The branching procedure defines a binary tree with nodes corresponding to sets of
repeaters "forced in" and "forced out" of the solution. Clearly, one of the nodes in the tree
corresponds to an optimal solution to the set covering the problem. If the optimal solution
to the set covering problem is not unique, then there are multiple nodes in the tree
corresponding to an optimal solution.
The basic strategy of the branch and bound procedure is to traverse the binary tree
until an optimal solution is obtained. The major difficulty with this approach is that it is
computationally expensive to traverse the entire tree. Hence, it is necessary to devise
mechanisms
- to eliminate branches of the tree
- to identify optimal solutions.
The practical mechanism to accomplish this is to compute upper and lower bounds for each
subproblem. An upper bound is computed by finding a feasible solution consisting of
- all repeaters "forced in" by the subproblem
- no repeaters "forced out" by the s' bproblem
- additional repeaters selected as necessary to meet the coverage constraints.
The procedure for computing the upper bounds described in Section 8.2.4.2.
Analogously, a lower bound is computed by a relaxed subproblem; this procedure is described
in Section 8.2.4.1.
Thus, given a new subproblem, a relaxed subproblem is solved to obtain a lower bound.
If this lower bound exceeds the current upper bound, then there is no possibility of finding a
node below the subproblem node with a better solution. Hence, the descendent nodes from
the current node need not be explored. In this case, the subproblem is said to fathom; this is
illustrated in Step 6 in the flow chart.
8.6
However, if the lower bound is less than the upper bound, then there Is potential for
finding a better solution if the tree below the current node is explored. Hence, branching is
performed to define two new subproblems. The algorithm continues until all subproblems
have been explored (which either fathom or produce a new pair of subproblems); the optimal
solution then is the feasible solution corresponding to the current upper bound.
8.2.4 Relaxed Subproblem
8.2.4.1 Lower Bound
The lower bound is computed by allowing fractions of a repeater to cover a terminal.
In Figure 8.3, the Set Covering Problem, P1, and Relaxed Set Covering, P2, are
mathematically defined. Given any feasible solution to the P1, we can find a
- feasible solution to P2
- the corresponding solution to P2 will have a smaller objective function value in
P2 than the feasible solution in P1.
Hence, the optimal objective value in P2 is a lower bound on the optimal objective value in
P1.
Solution of the Relaxed Subproblem consists of first defining weights wii such that the
optimal objective value of P2 is as large as possible. Hence, the lower bound will be as tight
as possible. The procedure for doing this is heuristic and is described below. After the
weights are selected, an optimal solution is trivially obtained to the relaxed subproblem by
choosing for all i:
Xik = 1 for Xik = minimum{w
X = 0 otherwise.
I = ' repeater which have not been forced I
8.7'S! .
__aciPI: Set Covering Problem
minimize z Z yj - 1 if repeater i Is selected- 0 otherwise.
s.t.
Xj I 1 (all terminals are covered)j
X < Yj (terminal i can be covered by
repeater if and only if repeater J isin the solution)
Xtj, Yj e C O,1
Y = 0 for all repeaters currently "forced
out."
Y = 1 for all repeaters currently "forced
in."
P2: Relaxed Set Covering Problem
minimize z I Wj X~i- ij
j >
Xij € (0,1)
I - { set of repeaters not forced in or not forced out }
FIGURE 8.3: MATHEMATICAL REPRESENTATION OF SET COVERING PROBLEM
AND RELAXED SET COVERING PROBLEM
Inac_Lower Bound Procedure
1. Let Wj be the remaining weight associated with repeater J. Initially, Wj = I for
all J. Let Si be the number of uncovered terminals which can be covered byrepeater J. Initial S = £ C where C. is the element of the coverage matrixj iassociated with the incidence of terminal i and repeater J. In particular, C = 1if repeater j can cover terminal i and 0 otherwise. Set Vi equal to 0 if terminal i
is covered by a repeater forced into current solution; one otherwise.
2. If S = 0 for all j, go to step 7.j
W.3. Find the repeater, K, for which A = Fis minimum.
J
4. Set W the weight associated with the i - j incidence, = A for all uncovered
rows covered by repeater K.
5. Update W. and S. as follows
W = 1 - Wi.
i C V
(Thus, W is the remaining weight to be given out for repeater j as S. is the
number of as yet uncovered terminals which can be covered by repeater J.)
6. Return to step 2.
(At this point, weights have been assigned; find optimal solution.)
7. Select the smallest non-zero elements in each row. This is equivalent to adding
the values of Wk found in Step 3.
ci 6.9
niac_8. Since the solution must be an integer, we can tighten the bound further by
increasing the value in Step 7 to the next highest integer (e.g., if the value was
5.1, return a 6; if the value was 8, return 8).
8.2.4.2 Upper Bound
To obtain the upper bound, a feasible solution is derived from the current solution,
which defines constraints on repeaters "forced in" and "forced out." Clearly the objective
value associated with a feasible solution is an upper bound, i.e., by definition, an optimal has
an objective value less than or equal to the objective value of any feasible solution.
Upper Bound Procedure
This is similar, but not identical, to the lower bounding procedure. A feasible solution
is an upper bound. The best feasible solution seen so far is the tightest upper bound.
1. Set Vi = 1 for all terminals not covered by forced in repeaters; V. = 0
otherwise; S. = Z V. Cj 1 i ji*1
2. Find max S . Select this repeater.
3. Set Vi = 0 for all terminals covered by the repeater in 2.
4. SetS = Z Vi Cj i ij*
5. If any V. = I for some i, return to Step 2; otherwise, all terminals are covered
and repeaters selected (with "forced in" repeaters) comprise a feasible solution.
8.2.5 Enhancements
8.2.5.1 Further Dominance
After certain repeaters are forced in and forced out, the dominance routine may be
, reexecuted to identify additional redundance in the subproblems.
8.10
8.2.5.2 Reevaluation of Lower Bound
When a subproblem is extracted from the stack, the lower bound associated with the
farther node of the subproblem node could be compared with the current upper bound. It ispossible that the current upper bound asociated may be less than this lower bound, i.e., the
upper bound could have been computed since the father subproblem was analyzed. Thus thecurrent subproblem fathoms, i.e., it is no longer desirable to branch from it. Of couse,
additional data structures are required to maintain the lower bound.
8.2.5.3 Termination
The algorithm terminates when all subproblems fathom, i.e., it is not worthwhile to
perform further branching form the node, or subproblems have no decendants, i.e., if it is
not possible to perform branching. This guarantees an optimal solution, but from a practical
point of view, it may be desirable to terminate before the optimal solution is obtained. For
example, the difference between the current upper bound and smallest lower bound may beso small, that the additional computation (which could be substantial) is not profitable. Thus
an alternative termination criteria is stopped when the difference between the current
upper bound and the lower bound for all subproblems which have yet to fathom is less than a
prespecified threshold.
8.2.5.4 K Cover Problem
The 1-cover algorithm above can be extended to the K cover problem. However,
- The dominance criteria would have to be modified to account for the8 requirement terminals must be covered by K repeaters.
- The lower bound coverage constraint would be modified so that
£ Xi_ KJ ii
The feasible solution in the upper bound would require that terminals be covered
by K repeaters.
!, 8.11
4 -h 1 . . . . . - . . . . .. . l I I l . . . . .. . . . .. . . . . . [ l l l l H l =
.. . . . . . . .. . .. . . . . . . . . ' m n . . . . . .. . . . .
These modifications are relatively simple conceptually, but the increase in
eompuational requirements could become substantial. These issues have not yet been
addressed in detail.
8.12
0
APPENDEX A
LONGLEY RICE CONNECTIVIY ANALYSIS
A.1
iiaci_A.1 Introduction
This Appendix describes the detailed work done on PR network connectivity using theLongley Rice propagation model. First, a mean attenuation loss model was formulated. Inthis model, a link exists between two PRU's if and only if the mean attenuation loss between
them (calculated by the Longley Rice model) does not exceed the specified loss threshold. Alarge number of experiments was performed to evaluate network connectivity (using themean attenuation loss criteria) for loss thresholds of 150 dB, 155 dB, and 160 dB.
The methodology for these experiments and validation of the model are described inSections A.2 and A.3, respectively. Then the connectivity results are described inSection A.4. These results indicated that the PR networks were sparsely connected and that
a substantial number of dedicated repeaters would be required to satisfy the needlines.Specifically, it was determined that even when the threshold is increased to 160 dB,
typically well over one-half of the needlines cannot be satisfied without the introduction ofdedicated repeaters. If terminals from other subnetworks are allowed to function as
repeaters, then the number of needlines that can be satisfied without the introduction ofdedicated repeaters significantly increases. However, in the best case for C2, about one-third of the needlines are still not satisfied.
As a result, it was decided to employ a more refined version of the Longley Rice
model. This model, referred to as the random Longely Rice model, is described inSection A.5 and has the capability of considering (in a probabilistic fashion) the effects oflocation variability on attenuation. The results of this model indicated that there wassubstantially more connectivity than predicted by the mean loss criteria. These results aredescribed in Section A.6. Specifically, for the command control subnetwork with a 150-dBthreshold, more than 95% of the PRU's belong to a single connected subset and over 90% ofthe needlines are satisfied. For both the Field Artillery and Air Defense subnetworks with a150-dB threshold, there exists only one connected subset; hence, all needlines can besatisfied without the introduction of dedicated repeaters.
Because of the wide disparity in the connectivity predicted by the two models, therewas much uncertainty as to how much connectivity there would be if the PR net would bedeployed. This led to the decision to employ a LOS criteria in nearly all experiments.
It is noted that the decision not to employ the Longley Rice in determining the links isnot meant as a criticism of the model. It is simply the recognition of the reality that astate-of-the-art technology is not able to predict attenuation loss with sufficient accuracy.
A.2
A.2 Methodology to Determine Connected Subsets
A.2.1 Network Definition
The Longley Rice Model [6i calculates the attenuation loss between two PRU's, as
described in Section 2. If the attenuation loss is less than the specified threshold, a direct
link exists. In this manner, the PRU's that can communicate directly can be identified. This
then defines a network G(N, A) with a set of vertices, N, comprised of the PRU's, and a set
of arcs, A, comprised of the direct links. Sensitivity studies were performed using the
threshold value as a parameter. Hence, the network is a function of the threshold value.
A connected subset is a set of at least two PRU's in which each PRU can communicate
with every other PRU in the subset by some path. There does not necessarily exist a direct
link between each set of PRU's in a subset; however, some path does exist.
A.2.2 Algorithm Description (based on Goodman and Hedetniemi [s])
In order to determine the connected subsets, a network adjacency data structure
representation of the direct links in the subnetwork is created. This data structure is made
up of two arrays, ADJ(K) and NEXT(K). Let p equal the number of PRU's in the subnetwork.
For 1< K< p, ADJ(K) = 0. If J = NEXT(K) for 0 < K < p, then ADJ(J) is the first vertex
adjacent to vertex K. The next vertex adjacent to K is contained in ADJ (NEXT(J)), and so
on.
The connected subsets are computed by a Depth First Search (DFS). Once the adja-
cency data structure has been defined, the DFS is executed in order to find a tree. In the
case being discussed, a tree is a connected subset. This DFS routine is called repeatedly
until all the connected subsets in a particular subnetwork are found. In the DFS algorithm
the variable V is the current vertex that is being examined. The DFS algorithm proceeds by
taking some arbitrary vertex adjacent to V, say vertex W. If vertex W has not yet been
L; visited, the algorithm Iteratively finds some adjacent vertex of W. If, however, vertex W
has been visited, the algorithm tries some other vertex adjacent to V, always looking for
unvisited vertices to visit next. If no unvisited vertices adjacent to V are found, the
algorithm backs up to the vertex from which V was visited and tries other adjacent vertices.
A.3
nacIn order to remember what the location was when DFS is called, the current values of
V and W are stored in a pushdown store. The latest V and W are always at the bottom of the
pushdown store. After visiting all the possible vertices of the current vertex and finding
that they have been visited, the pushdown store will have the value of the last vertex. Thus
it is possible to backtrack and visit its adjacent vertices.
A.2.3 Longley Rice Link Model Application
The input parameters for the Longley Rice link module consist of three types as listed
below. Numbers in parentheses below are numerical values for the corresponding
parameter.
1. Parameters which are fixed for our studies:
- The transmission frequency (1,800 MHz)
- Dielectric constant (15)
- Ground conductivity (0.005 mho/m)
- Sea Level Refractivity (310 N units)
- Antenna Polarization (1, vertical).
The numbers in parentheses above are the numerical values to be used in the studies.
2. PRU parameters:
The PR units location coordinates in linear distances (kilometers from NB
reference point)
- Siting parameters
- Antenna heights.
A.4
3. Topography Parameter:
- Terrain Irregularity Parameter (A h) value.
The fixed parameters were supplied by CORADCOM and are defined by data
statements in the Longley Rice subroutine, while the packet radio unit parameters are
passed as calling arguments to the Longley Rice subroutine. For terminals, a siting
parameter of zero (random) will be used while for repeaters, a siting parameter of 1 (well
chosen) will be used. The PR units coordinates and antenna heights are determined from the
data base (the Scenario Generator passes these parameters to PRSIM).
The terrain irregularity parameter A h, an input parameter for the Longley Rice
model, characterizes the irregularity of ground topography between two communicating
nodes and must be computed each time the Longley Rice subroutine is executed. Since the
ground irregularity may vary significantly from one area to another of the network terrain,
a matrix of Ah values (Figure A.1) is defined corresponding to a set of cells that cover the
area.
The h value used by the link module to determine attenuation loss between any two
nodes N1 and N2 (Figure A.2) not in the same cell is determined by linear interpolation:
h ~ h I + h2 * + h3*-hNI,N2 =h
N+ 92Nt3
This gives a weighted average value of h N ,N " Each cell that line N1, N2 passes has
its h value weighted by the length L t ass through. The weighted sum Z h. i isaveraged over the total length Ek i hi
A.3 Validation of Connectivity Results
From the Longley Rice Model, results on attenuation loss in dB versus distance in km
for three A h values have been found. Figure A.3 illustrates attenuation loss vs. range for
both Longley Rice results, as well as for published results [4] . The Ah values used in the
A.5
L% n Go q~ c0 o 0o c 40w r. -l 1 j co Go 0 r%. c.
o 0o
0-44
fn) 0 to 9-4 In o V-4a "4 f-. Ch In .o r.. Go0
CV en ~ m~ g) CV) V) Cj.0 0
In In
Go 40 en) in C> Cj 4DN~ 01 to % 0o Go V
Ny CV) 04 n Go en I*
CDD
Ch 4m) en M0- * 2!: ILC.. n en Nn *n L e
V4 Ln .% W- j c C30n In C% 0% 0 l N ) "0en CV) V4 In In In CV)
In 0* 01 IC 0 040) 40) CV 'r q-4 N ri cc)
0IL
LIco at C>Nr%
01 In W) 40) u4idN 4 )) 40) N N V-4 N~ ca ac
U. U.
I- UL.
0l- In N. No 014 0o ...N) 40) 40% NC en N 0
00
in %C r. Rr Nw In li 00 1
- - In I
A.6.
INA'
c FIGURE A.2: SAMPLE Ah CALCULATION
A.7
ATTENTUAT ION
LOSS (dB)
27Q
&~h=520 m
1,800 MPH z
50 Ah=330 m1,80.0 MHz
Ah=l40 m
230. 1,800 MH z
210 IEEE-MOUNTAINS
2m-ANTENNA HGT.
1901,8Mz/ 2i.-ANTENNA HEIGHT
170.
ISIo130 _j
1 1 20 30 50 100 200 500 Range (kmn)
FIGURE A.3: PATH LOSS VS. RANGE
A.$
iiaci
Longley Rice model are the minimum (140m), average (330m), and maximum (520m) values
recommended by ECAC for the Fulda Gap terrain being considered for the PR
communications system.
The curves indicate that the loss predicted by the Longley Rice Model exceeds the
values indicated by the published results. For a low A h (140m), the Longley Rice predicted
values lie between the published results for hills and mountains; typically, the Longley Rice
value is about 5 dB greater than the published results for hills. The results for an averageAh (330m) in the Fulda Gap are about 8 dB greater in attenuation loss than those for
mountains. The results for a high Ah (520m) are between 15 and 20 dB higher in attenuation
loss than those for mountains. The general trend of the curves is seemingly quite similar to
that of the published results.
From Figure A.1, it is found that for a given attenuation loss and a specified range ofA h, the distance between two PRU's must lie within a certain range (in kilometers). This
data is summarized in Table A.1, entitled "Range for Communication." For example,
consider the first entry in the table which has an attenuation loss of 150 dB, a A h value in
the interval (140 to 330), and distances between 1 and 6.2 km. This means that for a 150-dB
threshold and a Ah of 140m, the maximum range of a PR is 6.2 km. Similarly, for a Ah of
330 m, the maximum range of a PRU is only 1 km. For values of A h between 140 and 330m, interpolation would have to be used. At a threshold of 160 dB and an average Ah value
of 330 m, the range of a PR is only 2.7 km. This is considerably less than the 5 to 6 km
initially assumed by MITRE in deploying the units. Also, at the maximum value of Ah (520m), the attenuation loss exceeds the 160-dB threshold at a distance of less than 1 km.
At a frequency of 1,080 MHz, the results have also been found for path loss vs. range.
The results found by the Longley Rice Model for a Ah of 140 m are very similar (up to 10
km) to those presented in [4 ] for hills. As illustrated in Figure A.4, the difference is
typically less than 2 dB. The results predicted by the Longley Rice model for a A h of 330 m
are also very similar those given in [4 ] for mountains, but the discrepancy is somewhat
larger, especially for shorter distance. As shown in Figure A.4 the discrepancy is typically
less than 3 dB. The attenuation loss is very much dependent on location. Hence, a
substantial part of this discrepancy is likely due to the fact that Longley Rice and the
published results are based on data from different locations.
However, after 10 kin, the results found by the Longley Rice model differ greatly from
those presented in [4 1.
A.9
TABLE A.I: RANGE FOR COMMUNICATION
Attenuation Loss A h Values Distance in km
150 dB 140 < A h < 330 1 < dist < 6.2
150 dB 330 < A h < 520 0 < ist < I
155 dB 140 < A h .< 330 1.5 < dist < 9.5
155 dB 330 < A h < 520 0 < dist < 1.5
160 dB 140 < A h < 330 2.7 < dist < 12.2
160 dB 330 < A h < 520 0 < dlist < 2.7
175 dB 140 < A h < 330 10 < dist < 30
175 dB 330 < A h < 520 3.5 < dist < 10
S
240
220
200 MOUNTAINS
160 &h 140
ANTENNA HEIGHT =2M
140 d
1 5 10 20 50
FIGURE A.4: PATH LOSS VS. RANGE AT 1080 MHZ
A.1 1
A.4 Connectivity Results
A.4.1 Command and Control Results
As described in Section 2, the Command and Control subnetwork consists of 36 PRUs
and 35 needlines. In the second deployment, three of the units are lost; so in the DL BRAVO
deployment, there are only 33 units. In the analysis of the connectivity in the C2 subnet-
work, sensitivity studies on the threshold value were performed for each deployment. In
each case, threshold values of 150 dB, 155 dB and 160 dB were used.
In general, for all three deployments, there is very little connectivity at 150 dB;
increasing the threshold to 160 dB reduces the number of isolated repeaters to
approximately one-third of all PRU's, but only about one-third of the 45 needlines are
satisfied.
A.4.1.1 Deployment 1
At a threshold of 150 dB, there are three subsets each with two PR units; thus there
are 30 isolated PRU's. Only one functional needline of the 45 required is satisfied within the
connected subsets. Increasing the threshold to 155 and 160 dB provides significant improve-
ment in connectivity, but there are still a substantial number of isolated PR's and a large
number of needlines not satisfied. Specifically, at a threshold of 155 dB, there are three
subsets: one has 5 PR units, one has 3 PR units, and one has 4 PR units. In this case there
are 24 isolated repeaters, and 6 functional needlines of the 45 required are satisfied. At 160
dB, there are four subsets: one has 12 PR units, two of them have 3 PR units, and one has 2
PR units. Even in this case, there are still 16 isolated repeaters, and only 11 of the 45
needlines are satisfied.
A.4.1.2 DL ALPHA Deployment
In the Defense Line Alpha deployment, there is more connectivity than in the initial
deployment, but there are still substantial numbers of isolated PR's and unsatisfied
needlines. At a threshold of 150 dB, there are three subsets: one with 7 PR units, one with
3 PR units, and one with 2 PR units. Only 6 functional needlines of the 45 are satisfied, and
there are 24 isolated units.
A 12
Ilac_At a threshold of 155 dB, there is a substantial improvement in connectivity resulting
in two connected subsets: one with 16 PR units and one with 2 PR units. Fifteen functional
needlines are satisfied. At 160 dB, there is a slight improvement in connectivity with three
subsets: one has 16 PR units and the other two have 3 PR units. The same 15 functional
needlines are satisfied. Thus, nearly two-thirds of the network is connected, but only about
one-third of the needlines are satisfied. within the connected subsets.
A.4.1.3 DL BRAVO Deployment
In the Defense Line BRAVO deployment, at a threshold of 150 dB, there is one
connected subset with 15 PR units and 14 functional needlines are satisfied. This is roughly
comparable to the connectivity in the initial deployment at a 160-dB threshold, indicating
that the PRU's are more closely grouped in this deployment. Then at 155 dB there is
slightly more connectivity; there are two subsets having 16 PR units, and 3 PR units with 15
functional needlines are satisfied. At 160 dB, there are four subsets: one subset has 16 PR
units, two have 3 PR units, and one has 2 PR units. Again, 15 functional needlines are
satisfied.
A.4.2 Field Artillery Studies
As described in Section 2, the Field Artillery subnetwork consists of 94 PRU's and 177
functional needlines. In the first deployment three of the units are lost, so in DL ALPHA
there are 91 PRU's and 174 needlines. In DL ALPHA, 18 more units are lost so that in DL
BRAVO there are 73 units and only 156 needlines.
In the connectivity analysis of the Field Artillery subnetwork, sensitivity studies on
the threshold value were performed for each deployment. In each case threshold values of
150 dB, 155 dB, and 160 dB were used.
For each deployment there is significant connectivity at the 150-dB threshold, with
about 40% of the PR units belonging to some connected subset. However, only a small
fraction of the needlines are satisfied (2 to 3% for the initial deployment and DL ALPHA
deployment, and 13% for the DL BRAVO deployment). Increasing the threshold to 160 dB
causes a substantial increase in connectivity. At this threshold over 80% of the PR units
belong to a connected subset, but lea than 40% of the needlines are satisfied. Possibly due
A.13
ii1ac..to the attrition loss incurred, the connectivity at DL BRAVO at the 160-dB threshold is less
than that in the initial deployment. In particular 11% fewer needlines are satisfied in DL
BRAVO than in the initial deployment at the 160-dB threshold.
A.4.2.1 Initial Deployment
Of the 94 Field Artillery PRU'p at a threshold of 150 dB, there are 13 connected
subsets. Of the PRU's, 38% are included in these subsets, while 58 PRU's are isolated.
Three of the required functional needlines are satisfied within the connected subsets; this is
less than 2% of the 177 required functional needlines. Increasing the threshold to 155 dB
provides improvement in connectivity and, when the threshold is raised to 160 dB, nearly all
PR units belong to some connected subset. Similarly, as we raise the threshold, there is a
significant increase in the number of functional needlines satisfied. Specifically, at a
threshold of 155 dB there are still only 13 connected subsets; however, 57% of the PRU's are
included in these subsets, while 40 PRU's are still isolated. Twenty-two needlines of the 177
required are satisfied; this is approximately 121%. At 160 dB there are 14 subsets, yet
931% of the PRU's are included in these subsets. Only 6 PRUIS are still isolated. Although
this subnetwork has most PRUs included in a connected subset, only 40% (70 out of 177)
needlines are satisfied.
A.4.2.2 DL ALPHA Deployment
For DL ALPHA at 150 dB, there is similar connectivity to the initial deployment. Yet
at 160 dB, there is a small decrease in the connectivity relative to the connectivity in the
initial deployment. At a threshold of 150 dB, there are 14 connected subsets; 38% of the 91
PRU's are included in these subsets and 56 PRU's are isolated. Only 5 needlines out of 174
are satisfied, just 3%. At 155 dB there are 17 connected subsets with 67% of the PRU's
included in them; only 30 PRU's are still isolated. Twenty needlines out of 174 are satisfied,
about 111% of the total. At 160 dB there are only 9 connected subsets, containing 89% of
the PRUs in these subsets. Only 10 of the PRU's are isolated. Fifty-nine out of 174
needlines are satisfied, about 34%.
A.4.2.3 DL BRAVO
In DL BRAVO at 150 dB, there Is slightly more connectivity than there was in the
other two deployments. Of the PRU's, 46% are included In the connected subsets, while In
A-14
n1ac.the other two deployments at this threshold only about 38% of the PRU's were included in
the subsets. There are only 9 subsets at 150 dB; out of 76 PRU's, 41 are isolated. Twenty-
one of the 156 needlines are satisfied by these subsets, about 13.5%, which is significantly
more than the 2% and 3% of the other two deployments. At 155 dB, 67% of the PRU's are
included in 12 subsets and 25 PRU's are isolated. Twenty-nine of the 156 needlines are
satisfied, about 18%. At 160 dB there is less connectivity than in the other two
deployments, only 84% of the PRU's are included in the 7 connected subsets, and 12 PRU's
are still isolated. Forty-five of the 156 needlines are satisfied, about 29%. The loss of 21
PRUs from the initial deployment to DL BRAVO has decreased the number of needlines
satisfied at 160 dB by about 11%.
A.4.3 Air Defense Artillery
As described in Section 2, the Air Defense Artillery subnetwork consists of 182 PRU's.
In the initial deployment, 8 units are lost so that in DL ALPHA there are only 174 PRU's. In
DL ALPHA, 23 units are lost, but 5 units, which were held in reserve in the DL ALPHA
scenario, are redeployed so that in DL BRAVO there are 156 PRU's. In the analysis of the
connectivity of the ADA subnetwork, sensitivity studies on the threshold values were
performed for each deployment. In each case threshold values of 150 and 160 dB were used.
The results are summarized in the discussion below.
It is interesting to note that in general as the number of PRUs decreases (due to
attrition), at a threshold of 150 dB the percentage of connectivity increases. Presumably
isolated PRUs close to the battle front are lost. Also, at a threshold of 160 dB, as the
number of PRU's decreases, the connectivity also decreases. At the 150-dB threshold,
about 60% of the PRU's belong to a connected subset, and when the threshold is increased to
160 dB, about 90% of the PRU's belong to connected subsets. The needlines have yet to be
analyzed; this will be done when the needline analysis procedure is automated.
A.4.3.1 Initial Deployment
At a threshold of 150 dB, there are 24 connected subsets made up of 97 PRU's out of a
total 182 PRU's; thus about 53% of the PRU's belong to a connected subset, while 85 PRU's
are isolated. At a threshold of 160-dB, there are 11 subsets consisting of 171 PRU's. One
of these subsets consists of 108 PRU's. Thus 94% of the PRUs belong to some connected
subset and 59% belong to the same connected subset.
A.15
A.4.3.2 DL ALPHA Deployment
At a threshold of 150 dB, there are 30 subsets of 92 PRU's out of 174. Sixty-two
PRU's are isolated. At a threshold of 160 dB, there are 15 subsets of 157 PRU's, about 90%
connectivity. One of the subsets consists of 50 PRU's, and there are only 17 isolated PRU's.
A.4.3.3 DL BRAVO Deployment
At a threshold of 150 dB, there are 107 PRU's out of 156 in 17 subsets; this is about
68% connectivity. One of the subsets consists of 39 PRU's and there are 49 isolated PRUs.
At 160 dB, there are 138 PRU'S in 9 connected subsets. One subset has 50 PRU's (32% of all
PRU's) and another has 58 PRU's (37% of all PRU's). Thus approximately 88% of all PRU's
belong to a connected subset and only 18 PRU's are isolated.
A.4.4 Whole Network
The entire network (consisting of the three subnetworks Command and Control, Field
Artillery, and Air Defense Artillery) contains a total of 297 PRU's in the initial deployment.
Eleven PRU's are lost in this deployment so that in DL ALPHA there are 286 PRU's. In DL
ALPHA, 42 PRU's are lost, and 5 are held in reserve and then redeployed in DL BRAVO so
that there is a total of 249 PRU's in DL BRAVO.
In the connectivity analysis of the network, sensitivity studies on the threshold values
were performed for each deployment with threshold values of 150 and 160dB. At both 150
and 160 dB, there were substantial improvements in connectivity, but a large number of
needlines were not satisfied. It was determined for both the initial deployment and the DLS ALPHA deployment that over 200 PRU's belong to the same connected subset for a 160-
dBthreshold, and about 97% of the PRUs belong to some connected subset. The complete
set of needlines were not analyzed, but for the initial deployment of C2, it was found that
18 of 45 needlines were satisfied with the introduction of repeaters from other subnets (an£ increase of 7 needlines). For the DL ALPHA deployment, there was an increase (from 15 to
34) in the number of needlines being satisfied. However, in DL BRAVO at 160 dB, the
improvements in connectivity were not as significant. In this case only 99 terminals were
contained in the largest subset. More detailed results are presented below.
A.16'I±
A.4.4.1 Initial Deployment
At a threshold of 150 dB, 240 PRU's out of 297 are included in 44 sub&t . This meansthat about 81% are included in some connected subset and that 37 PRU's are isolated. At
160 dB, 290 PRU's are included in 12 subsets; thus about 97.5% of the PRU's are contained in
some connected subset. One of these subsets consists of 205 PRU's, while another has 43
PRU's. Only 7 PRU's are isolated. Thus when the entire network is considered at 160 dB,
substantial network connectivity exists.
For the Command and Control network at a threshold of 150 dB, using PRU's fromother subnetworks as repeaters, there are 7 PRU's in 2 subsets, but 29 PRU's are isolated.
Without the repeaters, there are 6 PRU's in 3 subsets. Four out of 45 needlines are satisfied
with the use of repeaters, but without them only 1 needline is satisfied. At 160 dB with the
use of repeaters, 29 PRU's are included in 3 subsets and 18 out of 45 needlines are satisfied.
Only 7 PRU's are isolated. Without the use of repeaters, 20 PRU's are included in 4 subsets
and only 11 of the 45 needlines are satisfied. Thus even with this substantial improvement
in connectivity, a substantial number of dedicated repeaters may be required to satisfy the
needlines.
A.4.4.2 DL ALPHA Deployment
At a threshold of 150 dB, 219 out of a possible 286 PRU's are included in 62 connected
subsets; this is about 76% of the PRU's with 47 PRUs isolated. At 160 dB, 278 PRU's are
included in 9 subsets. One of these subsets has 239 PRU's in it and only 8 PRU's are
isolated. Hence, in this deployment, there is also substantial connectivity at 160 dB.For C2 at a threshold of 150 dB, with PRU's in other subnetworks acting as repeaters,
13 PRUs are included in 2 subsets and 23 PRU's are isolated. Ten out of the 45 needlines
are satisfied. Without the use of repeaters, there are 12 PRU's in 3 subsets and only 6 out of
the 45 needlines are satisfied. At 160 dB with the use of repeaters, there are 2 subsets, onewith 28 PRUs and one with 3 PRU's. Only 5 PRU's are isolated and 34 out of 45 needlines
are satisfied. Without the use of repeaters, there are three subsets consisting of 22 PRU's,
and only 15 of the 45 needlines are satisfied. Thus, in this case, using terminals from other
subnetworks as repeaters significantly increases the number of needlines that can be
satisfied without the introduction of dedicated repeaters.
A.1 7
11C_A.4.4.3 DL BRAVO Deployment
At 150 dB, 198 PRU's out of a possible 249 are included in 34 subsets, about 80%. One
of these subsets has 63 PRU's and 51 PRU's are isolated. At 160 dB, 240 PRU's are included
in 14 subsets and only 9 PRU's are isolated. However, only 99 PRU's are contained in the
largest connected subset, as opposed to over 200 in both of the previous deployments.
For the C2 subnetwork at a. threshold of 150 dB, with terminals from other
subnetworks acting as repeaters, there are 2 subsets: one with 15 PRU's and one with 2
PRU's. Out of the 33 PRU's, 16 PRU's are isolated and 14 needlines out of 42 are satisfied.
Without the use of repeaters, there is one subset of 15 PRU's and the same 14 needlines are
satisfied. At 160 dB with repeaters, there are three subsets of 25 PRU'S, so that only 8
PRU's are isolated. Sixteen out of 42 needlines are satisfied by these subsets. Without the
use of repeaters, there are 24 PRU's in 4 subsets and 15 needlines are satisfied. In this case
the use of terminals from other subnetworks (functioning as repeaters) only minimally
increases the number of needlines satisfied.
A.5 Methodology for Random Longley Rice Model
One of the deficiencies of the Longley Rice model is that it incorporates all the
topographic information into a single parameter, the interdecile range (Ah). Hence, there
are salient characteristics of the terrain, e.g., foliage, that are not explicitly captured by
the Longley Rice model. As a result, the measured attenuation of two pairs of points with
the same interdecile range can be markedly different. It has been empirically observed [3]
that this location variability is normally distributed. Typically, the standard deviation of
the distribution is approximately 20 dB. This standard deviation is quite large compared to
the assumed connectivity attenuation loss threshold of 150 dB. Hence, the random
SI algorithm for determining connectivity that accounts for this location variability has been
applied to determine connectivity.
As indicated above, the analysis of attenuation-loss measurement data has shown that
the attenuation loss variation due to location is normally distributed about the mean
attenuation loss predicted by the Longley Rice model. An expression to find the standard
deviation has been developed by Dr. Longley [ 7]:
a = 6 + 0.55(Ah/A) 1/ 2 - 0.004(Ah/X) dB (1)
A.1 8
('I
for h/ X 4,700, or
a L = 24.9 dB
for Ah/ k< 4,700, where A h is the interdecile range of the pair of PR U'S and the wavelength
= 0.167 m.
Given the mean attenuation loss from the Longley Rice model and the standard
deviation, a spread of three standard deviations about the average attenuation loss is found.
Then, using a random number generator, an attenuation loss value from the normal
distribution within the plus/minus 3 standard deviations is generated.
This value for the attenuated loss is compared to the specified threshold, assumed to
be 150 dB in these experiments. If the attenuation loss is less than the specified threshold, a
direct link exists. In this manner, the PRU's that can communicate are identified. This then
defines a network G(N,A) with a set of vertices, N, comprised of the PRUS and a set of
arcs, A, comprised of the direct links. The Depth First Search described above is then
applied to find the connected subsets. Given the connected subsets and the needlines in the
traffic data base, a computer program was written to determine automatically which of the
needlines were satisfied within each connected subset.
To generate a random attenuation loss within the plus/minus 3 standard deviation,
Composition Method [5] is applied to compute the random variable. Define S as a random
variable such that:
S = Y1 + Y2 + + n'
where Y1 , Y2 ' " "I" Yn are independent samples of the uniform variates is between 0 and 1,
which are generated based on some random seed. Then, for large n, the variable S is
approximately normally distributed with mean n/2 and variance n/12. Then, the variable X
defined by:
S -( F )
X = +
approximates the normal variable with mean and variance o3. by choosing n = 12, the(.square root is eliminated. Note the value of n truncates the distribution at + 6a limits.
A.1 9
*
nacm
Thus, X is only an approximation to a variate with a standard normal distribution within b ].
Subsequent experiments were run ignoring variates beyond L3 I and similar results were
obtained.
To apply the above results, the standard deviation is obtained from Equation ( 1),
which is an expression developed by Dr. Longley for the standard deviation of the
attenuation loss. The mean 11 is the mean attenuation loss obtained from the Longley Rice
model.
A.6 Results
In this section, the results of the connectivity experiments are described. A needline
defines communication requirement between two PRU's. If the communication requirement
is in only one direction (A to B or B to A), then the needline is referred to as a half-duplex
needline. If the requirement is in both directions (A to B and B to A), the needline is
referred to as a full duplex needline.
In comparing the results found by the random method to those found using a mean
attenuation loss, there is a substantial increase in the connectivity. Even when the
threshold is raised to 160 dB, the connectivity is much less than that found with the random
method with a threshold of 150 dB.
A.6.1 Corn maed and Control Results
With the random connectivity algorithm at a threshold of 150 dB, there is one subset
of 34 oF the 36 PRU's in the initial deployment. All of the 45 needlines are full-duplex and
41 are satisfied. At this threshold for this subnetwork and this deployment, several analyses
have been made for different random values. Each time, the same number of PRU's
* belonged to the connected subset, though they were not the same PRU's. This results from
the fact that the direct links are not the same for each run.
At 150 dB, using the mean attenuation loss algorithm to find connectivity, there are
only three subsets of 2 PRU's each, and only one needline of 45 is satisfied. By increasing
£ the threshold to 160 dB for the initial deployment, there are four subsets: one has 12 PRU's,
two have 3 PRU's, and one has 2 PRU's. In this ease, 16 PRU's are isolated and only 11
needlines are satisfied. Even by increasing the threshold to 160 dB, the connectivity is still
much less that that found when using the random method. Many more repeaters are
necessary in order to achieve the proper connectivity. The Main CP (PRU 1 in the data
u A.20
base), which is the PRU with the heaviest traffic, has very few of its needlines satisfied
within the connected subset. For example, the three Brigades (2,3,4) are all about 20 to25 km away, and so would need about 4 to 5 repeaters to be able to communicate. The
Cavalry Squad (15) is at least 40 to 50 km away from the Main CP(l) and also has arequirement; in this case, it could use some terminals as repeaters, but would probably still
need about 8 repeaters. The Field Artillery (70) cannot communicate with any of thenecessary PRU's. The closest of the PRU's which has a requirement to the Field Artillery
(87) is about 10 to 15 km away and would need 2 to 3 repeaters. Of couse, the required
number of repeaters depends on the detailed topography and terminal repeater siting.
The Division Artillery (26) can communicate with the Main CP(1) and with one other ofthe PRU's with which it needs to communicate. But, all the other PRU's it needs to
communicate with are scattered across the terrain. The Division Support Command (20) is
probably the only PRU that has all its needlines satisfied. All those that communicate with
it lie in the same area.
When terminals in the two other subnetworks are used as repeaters (both at 150 and160 dB), there is quite a bit more connectivity. However, a large number of the C2 needlies
are still not satisfied. It was determined for both the initial deployment and the DL ALPHA
deployment that over 200 PRUs out of 297 PRU's belong to the same connected subset for a160 dB threshold. About 97% of the PRU's belong to the same connected subset. For the
initial C2 deployment, 18 out of the 45 needlines were satisfied with the introduction of
repeaters (an increase of 7 needlines). This is still much less than the number of needlines
satisfied by the random method.
Similar results have been found for the other deployments of the C2 subnetwork.
Therefore, the results of DL ALPHA and DL BRAVO are not stated in as much detail. At
DL ALPHA, using the random algorithm, there is one subset of 35 PRUs out of 36, and 43
out of a possible 45 needlines are satisfied. Using the mean attenuation loss value, there are
three subsets of 7, 3, and 2 PRU's, and only six functional needlines of the 45 required are
satisfied.
In summary, for all three deployments, the results obtained by the random method
show substantially better connectivity than those found by mean attenuation loss.
A.6.2 Field Artillery Results
For the initial deployment at a threshold of 150 dB, using the random method, 94 outof a possible 94 PRU's are in one connected subset; therefore, all 177 needlines are satisfied.
A.21
For the mean attenuation loss algorithm, 45 PRUts are included in 13 subsets. This is only
about 38% of the PRU's, and only 3 needlines are satisfied. In comparison to 150 dB, this is
a substantial increase, but it is still much less than with the use of the random method.
For DL ALPHA, once again using the random method, there is 100% connectivity; 91
of the 91 PRU's are in one subset and all 174 needlines are satisfied. For mean attenuation
loss, 35 PRU's are included in 14 connected subsets and only 5 needlines are satisfied. For
DL BRAVO, there is also 100% connectivity using the mean attenuation algorithm; all 73
PRU's are part of the same subset and all 156 needlines are satisfied. For the mean
attenuation loss, there are 35 PR units in 9 subsets. The connectivity is this deployment is a
little better than for the other deployments, but still only 46% of the PRU's are included.
At 160 dB, there are 7 subsets of 61 PRU's and 45 needlines are satisfied.
It is quite obvious that the use of the random method demonstrates much greater
connectivity.
A.6.3 Air Defense Artillery
This subset consists of 182 PRU's. By using the random method for the initial
deployment, all 182 PRU's in the ADA subnetwork belong to one subset and all needlines are
satisfied. for the mean attenuation loss criteria, there are 24 subsets made up of 97 of the
182 PRU's. At 160 dB, there are 11 subsets consisting of 171 PRU's. This is quite good;
however, since they are dispersed among separate subsets, the needlines satisfied by this
method are still not as many as with the random method.
A.22
APPENDIX B
FEASIBILITY OF MOBILE SUBSCRIBER ENTRY SYSTEM
AND
PACKET RADIO SYSTEM TO SUPPORT ARMY DATA NEEDLINES
B.1
Ilacm
V
FEASIBILITY OFMOBILE SUBSCRIBER ENTRY SYSTEM
ANDPACKET RADIO SYSTEM
TOSUPPORT ARMY DATA NEED LINES
&
_ MSE/PR/1
ila
THE-PROBLEM
GIVEN:. ARMY DATA NEED LINES
MSE CONCEPTS
PACKET RADIO SYSTEM CONCEPTS
'flLTE&INLi
- CAN THE MSE SYSTEM SUPPORT THE DATA NEED LINES?
- CAN A PACKET RADIO SYSTEM SUPPORT THE DATA NEED LINES?
- HOW MUCH NARROW BAND VOICE CAN BE CARRIED BY A PACKETRADIO SYSTEM (AT 2.4KBPS OR 1.2KBPS) AND STILL SUPPORTTHE DATA NEED LINES
CONSTRAINTS:
- ONE MONTH TIME FRAME
- DATA NEED LINES NOT FULLY SUITABLE TO INVESTIGATE* BROADCAST/BUFFERED SYSTEMS
- MSE IN CONCEPT STAGE AND NOT FULLY DEFINED
S- MSE VOICE REQUIREMENTS LIMITED IN DETAIL
MSE/PR/2
MS DESRIPIO
* RADIOTELEPHONE SYSTEM:
- MOBILITY OF RADIO
- ADDRESSABILITY OF TELEPHONE
BASIC COMPONENTS*
- TERMINAL (MST)
- CENTRAL (MSC)
- ACCESS UNIT (AU)
ci, MSE/PR/3 >
nlac.
MOBILE SUBSCRIBER TERMINAL (MST)
USER INSTRUMENT
VOICE - DIALUP OR DAS
DATA INTERFACE - DIALUP OR DAS
* PRIME MOVER
- JEEP
S- SHELTER
- COMMAND TRACK
- ARMY AIRCRAFT
FULLY COMSEC SECURED
S
C
0MSE/PR/I4 _______
nlac.
MOBILE SUBSCRIBER CENTRAL (MSC)
* RANGE EXTENSION THRU RETRANSMISSION
0 INTERFACE WITH CORPS SYSTEM
0 COMSEC KEY DISTRIBUTION
0 SHELTER MOUNTED
* SET UP / TEAR DOWN - 30 MINUTES
,9
nac.
ACCESS UNIT (AU)
INTERFACE FOR TEL CALLS INTO MSE SYSTEM
* LOCATED WITH TEL SWBD
* COMSEC SECURED
!' S
U
o M~~~~ESEIPR/G..........b
LLI
- ()
I .~- U Z --
oz Li- OLJ
= I-- = u
0
u I
L1
MSE/PR/7
iiac.
I-c cCD LOCCK o0
C/ = = I,
cnI)LJ L
MSE/PR/8
ilac.
MOBILE SUBSCRIBER TERMINAL
RADIO
I SIGNALING
II DATAI I INTERFACE
MSE/PR/9
nlacm
U MSE/PR/1 0
nlac.
0
V)
MSE/PR/1 I
nlac.
LLI-
CD,
0-
C=,,
KMSE/PR/ 12-----
£1/W/3SW
x
C-)
x
xLI
.....
TYPES OF REQUIREMENTS(WITHIN DIVISION)
GROUP SWITCHING A CONTINUITY COMMEN
ID TIME RAIE (CONTINUOUS(C)(SECONDS) PS) OR INTERMITTANT(1))
A 0 + (R. T.) 1.2 C PR-? MSE-NO
A 0 + (R.T.) 1.2 1 PR-?,MSE-NO
C 0 + (R.T.) 16.0 C DEDICATED LINKS
C 0 + (R.T.) 16.0 I PR-?,MSE-NO
D 0 + (R.T.) 41+ - 1K CI DEDICATED LINKS
E 1 1.2 I PR,MSE-YES
-(F)* 2 - 5 2.4 ? DIV - CORP
BATCH SCHEDULED 1.2 - 2,4 1 MSE,PR-YES
1 15+ 1.2 1 MSE,PR-YES
1 15+ 32 1 PR,MSE-YES
-(G)* 7- 15 1.2- 32 1 PR, MSE-YES
* **F, G IDENTIFICATION ASSIGNED BY NAC
R.T. - REAL TIME
0
SUMMAc.MSE, PACKET RADIO RESOURCES
INTRA-DIVISION DATA REQUIREMENTS
MSE PACKET RADIO OMMENT
GROUP CHANNELS/BUSY HR KBP$(BiUSY HR)
A,CON. DEDICATED LINKS
A,INT 9 .3 SWITCHING/QUEUEING
- .06
C,INT 20 10.8 FEASIBILITYQUESTIONABLE
4-
C,CONT DEDICATED LINKS
GROUP D DEDICATED LINKS
-(F) NO REQUIREMENTS
GROUP E 48 39.5a
BATCH 8 18.6 UNIFORMLY SCHEDULEDOVER 16 HOUR DAY
I .7 .2
-(G) 1.8 1.3
TOTAL EXCLUDING RT: 58.5 59.8TOTAL INCLUDING RT: 87.5 70.7
NOTE:REQUIREMENTS DO NOT INCLUDE POSSIBLE TAIL CIRCUITS TERMINATINGWITHIN DIVISION; NEED LINES WITH UNKNOWN VOLUMES WERE OMITTED.
MSE/PR/15
VOICE REQUIREMENTS*
TYPE OF CALLS IN PROGRESSCIO NORMAL MODE ECCMM
DIRECT 35
(ONE HOP)
ONE INTERMEDIATE 21 34RELAY
(TWO HOP)
TWO INTERMEDIATE 14 16RELAYS
(THREE HOP)
TOTAL CALLS 70 50
TOTAL EFFECTIVE 119 116CHANNELS USED
*(SOURCE: "PERFORMANCE SPECIFICATION FOR THE MOBILE SUBSCRIBER£ CENTRAL", TT-BI-9201-0063, 30 APRIL 1978, DRAFT, JOINT
TACTICAL COMM. OFFICE)
I
de
inac.MSE VOICE CALLS
l c
AA
4,
NORMAL MODE
AA
E~fl21QIPTERMINAL
C, Q ELAY MSE/PR/17
iiac.
MSE SUBSYSTEM SPECIFICATIONS
MAXIMUM CONFIGURATION
9 MSC'S
400 MST'S
60 2-CHANNEL AU'S
20 4-CHANNEL AU'S
" COERAE.L 25 X 30 KM FOR DIVISION
* MSC CAPACITY UP TO 12 SIMULTANEOUS CALLS
a SUBSYSTEM CAPACITY: UP TO 70 SIMULTANEOUS CALLS
£
0 MSE/PR/1 8
nlac.
PACKET RADIO, MsE CORRESPONDENCES
PACKET RADIO TERMINAL (PRT): MST
PACKET RADIO STATION: MSC
PACKET RADIO TERMINALS + : AUCONTROL UNIT +STATIC SUBSYSTEMTERMINATION UNIT =(PRAU)
IN ANALYSIS PERFORMED, USE DIRECT ONE-TO-ONE REPLACEMENT
MSI
, 0 SE/PR/1 9
iiacI
MSE REQUIREMENTS(BUSY HOUR)
COVERAGE: 25 X 30KM, ASSUME ALL POINTS CAN INTERFERE
(NO SPACE DIVERSITY)
DATA BITS PER CHANNEL: 16KBPS
£ BANDWIDTH PER CHANNEL: 25KHz
TOTAL ONE HOP DATA: 99 CHANNELS(EXCLUDING R.T. REQUIREMENTS)
TOTAL ONE HOP VOICE: 119 CHANNELS
TOTAL NUMBER OF CHANNELS REQUIRED: 218
SYSTEM BANDWIDTH FOR DATA: 2.48MHz
SYSTEM BANDWIDTH FOR VOICE: 2.98MHz
TOTAL SYSTEM BANDWIDTH REQUIRED: 5.46MHz
S
B . MSE/PR/20
na c.
APPROACH TO MIXED VOICE/DATA REQUIREMENTS
COMBINE MSE AND PACKET RADIO APPROACHES
- MULTIPLE CHANNELS (FREQUENCIES OR CODES)
- BROADCAST USE OF CHANNELS
- SLOTTED ALOHA ON CHANNELS (30% EFFICIENCY)
- ASSUME SAME EFFICIENCY (BITS/HERTZ) FOR MSE AND PACKETRADIO
z
8'
0 MSE/PR/21
riacm
SUMMARY REQUIREMENTSPLUS OVERHEADS
PACKET RADIO FOR DATA REQUIREMENTS(INTRA DIVISION, NON REAL TIME)
• TOTAL INFORMATION TRAFFIC PER BUSY HOUR: 59.8KBPS
AREA COVERAGE: 25 X 30 KM
* OVERHEAD: 20% TOTAL TRAFFIC v 71.8KBPS
* EFFECTIVE NET DATA TRAFFIC (FOLLOWING VOICE PATTERNS)
- 122KBPS
PACKET RADIO FOR VOICE REQUIREMENTS(2.4KBPS VOICE RATE)
MAXIMUM CALLS IN PROGRESS: 70
TOTAL EFFECTIVE ONE-HOP VOICE CALLS: 119
* VOICE PACKETIZING: lOOMS (240 BITS), 96 BIT HEADERg
TOTAL PACKET VOICE LOAD: 400KBPS
MSE/PR/22
n1ac.
PACKET RADIO FOR VOICE
* VOICE DIGITIZATION RATE
- A. 2.4KBPS
- B. 1.2KBPS
* VOICE PACKETIZATION RATE: looms
H EADER OVYERHEAD (96 BITS)
- A. 40%
- B. 80%
* CHANNEL PROCEDURE: SLOTTED ALOHA
S EFFICIENCY: .3
* CODING EFFICIENCY: MSE (16KBPS PER 25KHz CHANNEL) =.64
EFFECTIVE ONE HOP VOICE CALLS: 119
* BANDWIDTH REQUIRED FOR PACKET VOICE
- A. 2.0811Hz
- B. 1.34MHz
(BANDWIDTH FOR MSE VOICE - 2.98MHz)
MSE/PR/23
nlac.
SUMMARY
MSE VERSUS COMBINED PACKET RADIO/MSE SYSTEM
BAND WI DTH REQU IRED
SE MSEIPR MSE/PRREQUIREMENTS (ZAiKBPS VOICE) (1.2KBPS-VOICE)
BASIC SYSTEM FOR DATA 2.49MHz .64MHz .64MHz(NO VOICE)
PLUS APPROPRIATE R.T.
REQUIREMENTS* 3.72MHz .75MHz .75MHz
BASIC SYSTEM FOR voicE 2.98MHz 2.08MHz 1.34MHz
COMBINED VOICE AND DATA 6.7MHz 2.83MHz 2.09MHz
*IN MSE, REAL TIME REQUIREMENTS HANDLED AS DEDICATED CHANNELS
MSE/PR/24
nlac.
AREAS FOR FURTHER STUDY
FULLER ANALYSIS OF REQUIREMENTS
* EXACT NATURE OF "REAL TIME" REQUIREMENTS
- MESSAGE FRAIING, BUFFERING
- TRUE SPEEDS OF SERVICE REQUIRED
* MULTIPLEXING OF DATA REQUIREMENTS ON MSE
CHANNEL ACCESS PROTOCOLS FOR PACKET VOICE CAN DO BETTER THAN
.3 SLOTTED ALOHA EFFICIENCY USED
CHANNEL CODING EFFICIENCY (16KBPS/25KHz) USED FOR PACKET RADIOCAN BE IMPROVED FOR WIDER BAND PACKET RADIO CHANNELS
RANGE/DATA RATE TRADEOFFS TO ACHIEVE SOME SPATIAL DIVERSITYFOR PACKET RADIO AND/OR MSE
DYNAMIC VERSUS STEADY STATE SYSTEM ANALYSES
DETAILED HARDWARE IMPACT ANALYSES
MSE/PR/25
AD-AO87 417 NETWORK ANALYSIS CORP GREAT NECK NY FIG 17/2.1PACKET RADIO DEPLOYMENT STUOY.(U)
UNCLASSIFIED FR.207.01-Rl NLKO79C06
Ap So N
L Jackson, B., aWorking Paper
2. Tobagi, F., antTerminal ProbTransactions o
3. Goodman, S.,Algorithms, Mi
66(11), Noveml
5. Kobayshi, HR
Methodology,
0 8 16. Longley, A.,
Irreglar Terri
7. Longley, A.,'Systems," OT
DIBUOGRAPHY
Jackson, B., and F. Owens, "Data Communications Scenario," MITRE Corporation,
Working Paper 13483, January 1979.
Tobegi, P., and L. Kleinrock, "Packet Switching in Radio Channels Part 11: The Hidden
Terminal Problem in Carrier Sense Multiple Access and Busy Tone Solution," IEEE
Transactions on Communications, 23(12), December 1975.
Goodman, S., and S. Hedetniemi, Introduction to the Design and Analysis of
Algorithms, McGraw Hill, New York, 1977.
Kahn, R., et al, "Advances in Packet Radio Technology," Proceedings of the IEEE,
66(11), November 1978.
Kobayashi, H., Modeling and Analysis: An Introduction to System Performance
Methodology, Addison - Wesley, Reading, Massachusetts, 1978.
Longley, A., and R. Rice, "Prediction of Tropospheric Radio Transmission Over
Irregular Terrrain - A Computer Method," ESSA Technical Report ERL 79 ITS 67, July
1968.
Longley, A., "Location Variability of Transmission Loss - Land Mobile and Broadcast
Systems," OT Report, 78-67.